Read-aloud: natural neural voice via local Piper TTS

Replace the browser's robotic Web Speech API (espeak on Chromium) as the
primary read-aloud path with server-side Piper neural TTS, served by petal
and kept fully offline on millenia next to Ollama.

- internal/tts: proxy short passages to Piper, transcode WAV -> mp3/opus via
  ffmpeg, content-addressed disk cache (instant re-taps). Each Piper instance
  loads one voice, so language routes to its own endpoint:{voice}. Unknown
  language -> 404 so the client falls back to Web Speech. UTF-8-safe truncation
  for multibyte (Chinese) text.
- config: TTS_ENDPOINT / TTS_ENDPOINT_ZH / TTS_VOICE_EN / TTS_VOICE_ZH /
  TTS_CACHE_DIR / TTS_TIMEOUT / TTS_AUDIO_FORMAT. Route mounts only when
  TTS_ENDPOINT is set; otherwise unchanged behavior.
- web/audio/speech.ts: speak() hits /api/tts first, falls back to Web Speech on
  any failure; rapid-tap-safe via a request token. Call sites unchanged.
- deploy/: Piper user systemd units (EN :5005, ZH :5006), setup script, README.

English (en_US-amy-medium) and Chinese (zh_CN-huayan-medium) are both live.

Claude-Session: https://claude.ai/code/session_016Yr6jELuRc7hyzYLccQKZd

.env.example

@@ -7,6 +7,9 @@ BASE_URL=http://localhost:8080
 # Database (SQLite, pure-Go modernc — no cgo)
 DATABASE_PATH=./data/petal.db
 
+# On-disk store for images pasted/dropped/inserted in the editor
+IMAGE_DIR=./data/images
+
 # LLM
 LLM_BACKEND=vllm                  # vllm | ollama
 LLM_ENDPOINT=http://localhost:8000   # vLLM :8000, Ollama :11434
@@ -14,6 +17,20 @@ LLM_MODEL=                        # checkpoint model (small/fast). Never hardcod
 LLM_CHAT_MODEL=                   # Ask Petal model (Mandarin-native). Falls back to LLM_MODEL if empty.
 LLM_TIMEOUT=30s
 
+# Read-aloud (TTS). Off unless TTS_ENDPOINT is set — when empty, the /api/tts route
+# is not mounted and the frontend falls back to the browser's Web Speech API.
+# Backed by a local Piper HTTP server (python3 -m piper.http_server).
+# Each Piper HTTP server loads ONE voice, so English and Chinese need separate
+# instances (different ports). zh is only routed when both TTS_ENDPOINT_ZH and
+# TTS_VOICE_ZH are set; otherwise Chinese falls back to Web Speech.
+TTS_ENDPOINT=                     # e.g. http://127.0.0.1:5005 — empty disables server TTS
+TTS_ENDPOINT_ZH=                  # e.g. http://127.0.0.1:5006 — Chinese Piper instance
+TTS_VOICE_EN=en_US-amy-medium     # Piper voice id for English
+TTS_VOICE_ZH=zh_CN-huayan-medium  # Piper voice id for Chinese
+TTS_CACHE_DIR=./data/tts          # on-disk store for synthesized clips (content-addressed)
+TTS_TIMEOUT=15s
+TTS_AUDIO_FORMAT=mp3              # mp3 | opus | wav — mp3/opus transcode Piper's WAV via ffmpeg
+
 # --- Deferred (not wired in the local-dev build) ---
 
 # Auth (Authentik OIDC) — deferred; single hardcoded local user for now

BUILD_PLAN.md

@@ -95,6 +95,16 @@ Multi-session build. **Source of truth for what's done and what's next.** Update
 - Tests: `tags_test.go` (full lifecycle — create/idempotent/color-coerce/rename-recolor/assign/unassign/doc-list inclusion/roster counts/delete-cascade/404s), `search_test.go` (EN FTS, CJK FTS, 2-char CJK LIKE fallback, title-only, case-insensitive, empty/no-match, **edit re-indexes via the update trigger**). go build/vet/test clean, tsc clean, vite build OK. Live smoke vs the binary on a throwaway DB (port 8061, LLM pointed at a dead host): search EN/CJK/2-char all highlighted, tag create+assign+roster-counts+doc-list-tags+delete-cascade, check→502 (warm path), new CSS classes (`petal-tag-chip`/`petal-scrim`/`petal-drawer-open`/`pointer:coarse`) and the 小助手在休息 string present in the served bundle. FTS backfill of pre-existing docs verified separately.
 - **Known limitations**: trigram FTS snippets/ranking treat the query as a contiguous phrase (multi-term relevance is substring, not BM25-per-term) — fine for a personal corpus. Search is title+body only (not tag names). Hover gloss and right-click word lookup remain pointer-oriented (long-press contextmenu on touch is browser-dependent); the spelling/suggestion cards and rewrite bubble are fully touch-reachable.
 
+### Phase 11 — Writer power-ups ✅
+- [x] **In-document Find & Replace (Ctrl/Cmd+F)** — `SearchHighlight` ProseMirror extension (decorations, not marks — same anchoring discipline as the suggestion/spell layers; matches recomputed per-textblock on every edit, never stranded). `FindReplace` bar (bilingual zh-first): live match highlighting, ↑/↓ step-through with no-selection DOM scroll-into-view (so it never pops the rewrite bubble), match-case toggle, replace / replace-all (replace-all applies back-to-front so earlier edits don't shift later positions). Honey wash on all matches, rose ring on the active one.
+- [x] **Read-aloud / TTS** (`web/src/audio/speech.ts`) — Web Speech API, offline, feature-detected. 🔊 in the `WordCard` (pronounce the word) and the selection bubble (read the selection). Pairs with the phonetic line.
+- [x] **Keyboard + touch access to the ESL helpers** — refactored the right-click word-lookup into a position-based `openWordLookup(pos)`; now also driven by **Ctrl/Cmd+D** (look up the word at the caret) and a **touch long-press** (~500ms, the touch equivalent of right-click). **Ctrl/Cmd+J** rewrites the selection more naturally. (Closes the "pointer-only" limitation noted in Phase 10.)
+- [x] **Whole-corpus backup** — `GET /api/docs/export-all?format=md|docx|…` zips every doc (reuses the per-doc renderers; de-duplicates same-titled filenames; dated `petal-backup-YYYY-MM-DD.zip`). Static route takes priority over `/{id}` in chi — covered by `TestExportAll`. Sidebar footer "备份 · Back up all: Word / Markdown" download links.
+- [x] **Smart typography** (`Typography.ts`) — dependency-free input rules: curly quotes, em-dash (`--`), ellipsis (`...`). ASCII-only triggers so CJK fullwidth punctuation is untouched; every rule is plain-Undo-able.
+- [x] **Org niceties** — duplicate-a-document (App `handleDuplicate` → "… (副本)", copies body/tone, not tags), sidebar **sort** (Recent / Title / Longest), and a **document outline** popover in the toolbar (headings → click to scroll, indented by level).
+- [x] **English phonetic (pivot from pinyin)** — for a native-Mandarin English learner the useful pronunciation aid is the English IPA, not pinyin (she reads Chinese fluently). `scripts/build_phonetic.py` extracts ECDICT's `phonetic` column (same source/freq-gate as the gloss); `phonetic.json.gz` embedded + lazily loaded; `Result.Phonetic` resolved via the same de-inflecting candidate walk; shown as `/ˈrɪvər/` in the WordCard. **Full dataset built from ECDICT: 46,579 words (361KB gz)**, in line with the other lexicon assets. The script also has a `--seed` mode (71 curated common words) that ships as a fallback / works without the csv. Curated seed entries (clean IPA) override the ECDICT form where both exist.
+- Verified: go build/vet/test (incl. new `TestExportAll`), tsc, vite build all clean; live smoke vs throwaway binaries — `/word/river|running|serendipity|rivers` all return phonetic (`rivers`→`river` de-inflected), export-all returns a valid 2-entry zip with de-duped CJK filenames + dated name, route doesn't collide with `/{id}`.
+
 ### Deferred (post-v1-local)
 - [ ] Authentik OIDC auth + session middleware  ← **on hold: user doing foundational work first**
 - [ ] Copyleaks Tier-2 + webhook HMAC
@@ -105,6 +115,7 @@ Multi-session build. **Source of truth for what's done and what's next.** Update
 - [x] **Phase 10 — organization & polish**: cross-doc search, tags, tablet/touch polish, warm LLM-down failure states. ✅ (see Phase 10 above; "tags only" chosen over folders, FTS5 over LIKE)
 
 ## Session log
+- 2026-06-26: **Phase 11 complete** (writer power-ups, batch requested as "do it all"). Seven features: (1) in-doc **Find & Replace** — `SearchHighlight` decoration extension + `FindReplace` bar (Ctrl/Cmd+F, match-case, replace-all back-to-front, DOM scroll that doesn't trigger the selection bubble); (2) **read-aloud** Web Speech util + 🔊 in WordCard & selection bubble; (3) **keyboard/touch access** — Ctrl/Cmd+D caret lookup, Ctrl/Cmd+J rewrite, touch long-press (refactored `handleContextMenu` → shared `openWordLookup(pos)`); (4) **export-all** backup zip (`GET /api/docs/export-all`, `TestExportAll`, sidebar download links); (5) **smart typography** input-rules extension (curly quotes/em-dash/ellipsis, ASCII-only so CJK untouched); (6) **duplicate doc + sidebar sort + outline popover**; (7) **English phonetic** (pivoted from pinyin — IPA is what an English learner needs; pinyin annotates Chinese she already reads) via `scripts/build_phonetic.py` + embedded `phonetic.json.gz` + `Result.Phonetic` + WordCard `/ˈrɪvər/` line — **full 46,579-word dataset built from ECDICT** (the csv re-download worked; `--seed` mode kept as a csv-free fallback). Also folded in this session: the **selection-bubble vs copy/paste fix** (bubble deferred to pointer-up + container `pointer-events:none` so it never sits where you click). go build/vet/test + tsc + vite all clean; live smoke verified word-phonetic (incl. de-inflection) + export-all zip (de-duped CJK names, route priority). Next: deferred bucket (auth/Copyleaks/deploy), still on hold per user.
 - 2026-06-26: **Phase 10 complete** (organization & polish). Scope confirmed with user: all four areas, **tags** (not folders), **FTS5** search. Backend: migration `0004_tags_and_search` (tags + document_tags + `documents_fts` trigram virtual table with sync triggers + back-fill); `db.Tag` model + color constants; `internal/docs/tags.go` (tag CRUD + idempotent assignment + `tagsByDoc` helper, doc list now carries tags); `internal/docs/search.go` (`GET /api/search`, FTS for ≥3 runes + LIKE fallback for 1-2, Go-built sentinel-highlighted rune-aware snippets, owner-scoped). Mounted `/api/tags` + `/api/search` in main.go. Frontend: `useTags`, `TagChip`/`TagPicker`/`SearchBox`, rewritten `DocList`/`DocListItem` (chips + filter bar + search), `api.search`/tag methods + `splitSnippet`/`tagColorVar`; responsive sidebar drawer (hamburger + scrim, <768px) + `pointer:coarse` tap-target/affordance CSS; tap-to-open + outside-pointerdown-close for suggestion cards (touch); `useCheckpoint` `llmDown` flag → warm bilingual "小助手在休息" StatusBar note. Tests: `tags_test.go`, `search_test.go` (incl. update-trigger re-index). All builds/tests/vet/tsc/vite clean; live smoke vs binary on :8061 (dead LLM host) verified search EN/CJK/2-char, full tag lifecycle, check→502 warm path, bundle contents; FTS backfill of pre-existing docs verified. **All v1 phases (0–7) + post-v1 product (8–10) done.** Remaining: deferred bucket (auth/Copyleaks/deploy), on hold per user.
 - 2026-06-26: **Phase 9 complete** (ESL superpowers: inline Chinese gloss + tone-rewrite). Decisions confirmed with user: gloss is an **offline EC dictionary** (instant, LLM-down-proof, fits the embedded-lexicon ethos), rewrite is a **selection bubble**. Data: `scripts/build_gloss.py` builds `internal/lexicon/data/gloss.json.gz` from ECDICT (66MB csv → 1.3MB gz, 57k freq-≤50k words, cleaned/trimmed). Backend: `lexicon` gloss map + `Gloss()`/`Result.Gloss` + `GET /api/gloss/{word}`; `llm.RunRewrite` + rewrite prompt/`styleGuidance`; `internal/suggestions/rewrite.go` (`POST /api/docs/:id/rewrite`, stateless, owner-scoped). Frontend: `GlossTip` hover tooltip (350ms delay, reuses `wordAt`, CJK-safe) + gloss line in `WordCard`; `SelectionBubble` + `RewritePreview` wired through `EditorCore` (onMouseMove/onSelectionUpdate, request-token guards, clears on edit/doc-switch); `api.glossWord`/`api.rewriteSelection`; CSS for the three new surfaces (+ print-hidden). Tests added in `lexicon` and `suggestions`. All builds/tests/vet/tsc/vite clean; live smoke vs fake vLLM on :8055 verified gloss + rewrite + 400/404/502 paths. Next: **Phase 10 (organization & polish).**
 - 2026-06-26: **Phase 8 complete** (Trust foundation: version history + export) + empty-doc fix. Backend: migration `0003_document_versions`; `internal/docs/versions.go` (throttled auto-snapshot wired into `update`, manual snapshot, restore-with-pre_restore, prune to 40 auto, owner-scoped via join) and `internal/docs/export.go` (pure-Go Tiptap-JSON → md/html/txt/docx, no deps, CJK-safe filenames via RFC 5987). `DocumentVersion` model + kind constants. Tests: `versions_test.go`, `export_test.go` (incl. valid-zip docx assertion). Frontend: `api.client` version/export methods; `ExportMenu` + `HistoryPanel` components wired into the title row; `@media print` stylesheet + `.petal-no-print` for the browser PDF path; `editorEpoch` remount on restore. Empty-doc fix in `App.tsx` (blank drafts reuse-on-create + discard-on-leave via refs to dodge stale closures); deleted 2 orphan empties from the live :8099 DB. Multi-session plan agreed: this session = Phase 8; **Phase 9 (ESL gloss + tone-rewrite)** next, then **Phase 10 (search/folders/polish)**; **auth/deploy (was Phase 11) shelved** until user's foundational work lands. All builds/tests/vet clean; live smoke verified the full version+export+restore flow end-to-end. Next: **Phase 9.**

cmd/server/main.go

@@ -15,9 +15,11 @@ import (
 	"gitea.parodia.dev/drwily/petal/internal/config"
 	"gitea.parodia.dev/drwily/petal/internal/db"
 	"gitea.parodia.dev/drwily/petal/internal/docs"
+	"gitea.parodia.dev/drwily/petal/internal/images"
 	"gitea.parodia.dev/drwily/petal/internal/lexicon"
 	"gitea.parodia.dev/drwily/petal/internal/llm"
 	"gitea.parodia.dev/drwily/petal/internal/suggestions"
+	"gitea.parodia.dev/drwily/petal/internal/tts"
 	"gitea.parodia.dev/drwily/petal/web"
 )
 
@@ -80,6 +82,21 @@ func main() {
 		lex := lexicon.NewHandler()
 		api.Mount("/word", lex.Routes())
 		api.Mount("/gloss", lex.GlossRoutes())
+
+		// Editor image uploads, stored on disk and served back by content hash.
+		imgHandler, err := images.New(cfg.ImageDir)
+		if err != nil {
+			log.Fatalf("image store: %v", err)
+		}
+		api.Mount("/images", imgHandler.Routes())
+
+		// Read-aloud: proxy short passages to a local Piper TTS server. Only
+		// mounted when TTS_ENDPOINT is configured; otherwise the frontend falls
+		// back to the browser's Web Speech API on its own.
+		if ttsHandler, ok := tts.New(cfg); ok {
+			api.Mount("/tts", ttsHandler.Routes())
+			log.Printf("read-aloud enabled (TTS endpoint=%s)", cfg.TTSEndpoint)
+		}
 	})
 
 	// Everything else: serve the embedded SPA (with index.html fallback for client routing).

deploy/README.md

@@ -0,0 +1,65 @@
+# Deploying read-aloud (Piper TTS) to millenia
+
+Petal's read-aloud generates audio with a local **Piper** neural-TTS server and
+transcodes it to mp3 with **ffmpeg**. Both run on millenia (192.168.1.212);
+nothing leaves the box. If Piper is down or `TTS_ENDPOINT` is unset, the frontend
+falls back to the browser's Web Speech API automatically.
+
+## 1. Piper as a systemd service (one-time)
+
+```bash
+# from this repo, on your workstation:
+scp deploy/piper.service deploy/setup-piper.sh 192.168.1.212:/tmp/
+ssh 192.168.1.212 'cd /tmp && sudo ./setup-piper.sh'
+```
+
+`setup-piper.sh` creates `~/piper/venv`, installs `piper-tts[http]`, downloads the
+`en_US-amy-medium` voice into `~/piper/voices`, installs+enables `piper.service`
+(loopback :5005), and smoke-tests it. Idempotent.
+
+Check it any time: `systemctl status piper`, `journalctl -u piper -f`.
+
+## 2. Point petal at Piper + redeploy the binary
+
+Add to petal's environment (its `.env` or launch env):
+
+```
+TTS_ENDPOINT=http://127.0.0.1:5005
+TTS_VOICE_EN=en_US-amy-medium
+TTS_AUDIO_FORMAT=mp3
+```
+
+Then ship the rebuilt binary (`go build -o petal ./cmd/server` already done) and
+restart the petal `:8088` session. Confirm the log line:
+`read-aloud enabled (TTS endpoint=http://127.0.0.1:5005)`.
+
+## 3. Verify
+
+```bash
+# on millenia — end-to-end through petal, including ffmpeg transcode:
+curl -sf -X POST localhost:8088/api/tts \
+  -H 'Content-Type: application/json' \
+  -d '{"text":"hello there","lang":"en-US"}' -o /tmp/petal-tts.mp3 \
+  && file /tmp/petal-tts.mp3   # expect: Audio file ... MPEG ... layer III
+```
+
+- Second identical call is served from the cache (`~/petal/.../data/tts/*.mp3`).
+- A language with no configured instance returns 404 → client uses Web Speech.
+- Browser check via the uitest harness (`~/petal/uitest`): tap a word in the
+  WordCard / select a sentence and hit speak — expect the natural Piper voice.
+
+## Chinese voice (live)
+
+Each Piper HTTP server loads ONE model, so Chinese runs as a **second instance**:
+`piper-zh.service` on :5006 with `zh_CN-huayan-medium`. Deployed via:
+
+```bash
+scp deploy/piper-zh.service 192.168.1.212:~/.config/systemd/user/
+ssh 192.168.1.212 'export XDG_RUNTIME_DIR=/run/user/$(id -u)
+  ~/piper/venv/bin/python -m piper.download_voices zh_CN-huayan-medium --data-dir ~/piper/voices
+  systemctl --user daemon-reload && systemctl --user enable --now piper-zh.service'
+```
+
+Then in petal's `start.sh`: `TTS_ENDPOINT_ZH=http://127.0.0.1:5006` and
+`TTS_VOICE_ZH=zh_CN-huayan-medium`. The handler maps language → instance from config,
+so adding more languages is just another instance + env pair (no code change).

deploy/piper-zh.service

@@ -0,0 +1,17 @@
+[Unit]
+Description=Piper TTS HTTP server — Chinese voice (read-aloud backend for petal)
+After=network-online.target
+Wants=network-online.target
+
+[Service]
+Type=simple
+# Bound to loopback only — petal proxies to it; never exposed off-box.
+ExecStart=%h/piper/venv/bin/python -m piper.http_server \
+  -m zh_CN-huayan-medium \
+  --data-dir %h/piper/voices \
+  --host 127.0.0.1 --port 5006
+Restart=on-failure
+RestartSec=3
+
+[Install]
+WantedBy=default.target

deploy/piper.service

@@ -0,0 +1,17 @@
+[Unit]
+Description=Piper TTS HTTP server (read-aloud backend for petal)
+After=network-online.target
+Wants=network-online.target
+
+[Service]
+Type=simple
+# Bound to loopback only — petal proxies to it; never exposed off-box.
+ExecStart=%h/piper/venv/bin/python -m piper.http_server \
+  -m en_US-amy-medium \
+  --data-dir %h/piper/voices \
+  --host 127.0.0.1 --port 5005
+Restart=on-failure
+RestartSec=3
+
+[Install]
+WantedBy=default.target

deploy/setup-piper.sh

@@ -0,0 +1,43 @@
+#!/usr/bin/env bash
+# One-time Piper TTS setup on millenia as a *user* systemd service (no sudo).
+# Idempotent — safe to re-run. Run on the box:
+#   bash setup-piper.sh
+# To survive logout/reboot, also (once, needs root):
+#   sudo loginctl enable-linger "$USER"
+set -euo pipefail
+
+PIPER_HOME="$HOME/piper"
+VENV="$PIPER_HOME/venv"
+VOICES="$PIPER_HOME/voices"
+VOICE=en_US-amy-medium
+UNIT_DIR="$HOME/.config/systemd/user"
+export XDG_RUNTIME_DIR="/run/user/$(id -u)"
+
+echo ">> ffmpeg check (petal transcodes Piper WAV -> mp3 with it):"
+command -v ffmpeg >/dev/null || { echo "ffmpeg not found on PATH"; exit 1; }
+
+echo ">> venv + piper-tts[http]"
+[ -d "$VENV" ] || python3 -m venv "$VENV"
+"$VENV/bin/pip" install --upgrade pip >/dev/null
+"$VENV/bin/pip" install "piper-tts[http]"
+
+echo ">> download voice $VOICE"
+mkdir -p "$VOICES"
+"$VENV/bin/python" -m piper.download_voices "$VOICE" --data-dir "$VOICES"
+
+echo ">> install user service"
+mkdir -p "$UNIT_DIR"
+install -m 0644 "$(dirname "$0")/piper.service" "$UNIT_DIR/piper.service"
+systemctl --user daemon-reload
+systemctl --user enable --now piper.service
+
+echo ">> wait + status"
+sleep 2
+systemctl --user --no-pager --full status piper.service | head -n 6 || true
+
+echo ">> smoke test"
+curl -sf -X POST localhost:5005/ \
+  -H 'Content-Type: application/json' \
+  -d "{\"text\":\"hello there\",\"voice\":\"$VOICE\"}" -o /tmp/piper-test.wav \
+  && echo "OK: /tmp/piper-test.wav ($(stat -c%s /tmp/piper-test.wav) bytes)" \
+  || { echo "smoke test FAILED"; journalctl --user -u piper.service --no-pager | tail -n 20; exit 1; }

internal/config/config.go

@@ -11,6 +11,7 @@ type Config struct {
 	Port         string
 	BaseURL      string
 	DatabasePath string
+	ImageDir     string // on-disk store for editor image uploads
 
 	// LLM
 	LLMBackend   string // "vllm" | "ollama"
@@ -19,6 +20,17 @@ type Config struct {
 	LLMChatModel string // Ask Petal model; falls back to LLMModel if empty
 	LLMTimeout   time.Duration
 
+	// TTS (read-aloud). Off unless TTSEndpoint is set — when empty, the /api/tts
+	// route isn't mounted and the frontend falls back to the browser's Web Speech
+	// API. Endpoint points at a local Piper HTTP server.
+	TTSEndpoint   string // Piper instance serving the English voice
+	TTSEndpointZH string // Piper instance serving the Chinese voice; empty = zh falls back to Web Speech
+	TTSVoiceEN    string // Piper voice id for English (e.g. en_US-amy-medium)
+	TTSVoiceZH    string // Piper voice id for Chinese (e.g. zh_CN-huayan-medium)
+	TTSCacheDir string // on-disk store for synthesized clips (content-addressed)
+	TTSTimeout  time.Duration
+	TTSFormat   string // mp3 | opus | wav — mp3/opus transcode Piper's WAV via ffmpeg
+
 	// Auth (deferred — not wired in the local-dev build, kept for later)
 	AuthentikURL          string
 	AuthentikClientID     string
@@ -32,6 +44,7 @@ func Load() *Config {
 		Port:         env("PORT", "8080"),
 		BaseURL:      env("BASE_URL", "http://localhost:8080"),
 		DatabasePath: env("DATABASE_PATH", "./data/petal.db"),
+		ImageDir:     env("IMAGE_DIR", "./data/images"),
 
 		LLMBackend:   env("LLM_BACKEND", "vllm"),
 		LLMEndpoint:  env("LLM_ENDPOINT", "http://localhost:8000"),
@@ -39,6 +52,14 @@ func Load() *Config {
 		LLMChatModel: env("LLM_CHAT_MODEL", ""),
 		LLMTimeout:   envDuration("LLM_TIMEOUT", 30*time.Second),
 
+		TTSEndpoint:   env("TTS_ENDPOINT", ""),
+		TTSEndpointZH: env("TTS_ENDPOINT_ZH", ""),
+		TTSVoiceEN:    env("TTS_VOICE_EN", "en_US-amy-medium"),
+		TTSVoiceZH:    env("TTS_VOICE_ZH", "zh_CN-huayan-medium"),
+		TTSCacheDir: env("TTS_CACHE_DIR", "./data/tts"),
+		TTSTimeout:  envDuration("TTS_TIMEOUT", 15*time.Second),
+		TTSFormat:   env("TTS_AUDIO_FORMAT", "mp3"),
+
 		AuthentikURL:          env("AUTHENTIK_URL", ""),
 		AuthentikClientID:     env("AUTHENTIK_CLIENT_ID", ""),
 		AuthentikClientSecret: env("AUTHENTIK_CLIENT_SECRET", ""),

internal/docs/export.go

@@ -9,15 +9,101 @@ import (
 	"fmt"
 	"net/http"
 	"strings"
+	"time"
 
 	"github.com/go-chi/chi/v5"
 
 	"gitea.parodia.dev/drwily/petal/internal/db"
 )
 
-// exportRoutes registers the download endpoint: GET /api/docs/{id}/export?format=md|html|txt|docx
+// exportRoutes registers the download endpoints: one document
+// (GET /api/docs/{id}/export?format=…) and a whole-corpus backup zip
+// (GET /api/docs/export-all?format=…). The static "export-all" segment takes
+// priority over the {id} param in chi's router, so the two don't collide.
 func (h *Handler) exportRoutes(r chi.Router) {
 	r.Get("/{id}/export", h.export)
+	r.Get("/export-all", h.exportAll)
+}
+
+// exportAll streams every document the user owns, each rendered in the requested
+// format, bundled into a single zip — a one-click "download all my writing"
+// backup. Per-doc version history guards against bad edits; this guards against
+// a lost disk. Reuses the same renderers as the single-document export.
+func (h *Handler) exportAll(w http.ResponseWriter, r *http.Request) {
+	formatKey := r.URL.Query().Get("format")
+	if formatKey == "" {
+		formatKey = "md"
+	}
+	format, ok := exportFormats[formatKey]
+	if !ok {
+		badRequest(w, "unsupported export format")
+		return
+	}
+
+	rows, err := h.DB.Query(
+		`SELECT id, user_id, title, content, content_text, tone, word_count, created_at, updated_at
+		   FROM documents
+		  WHERE user_id = ?
+		  ORDER BY updated_at DESC`,
+		db.LocalUserID,
+	)
+	if err != nil {
+		serverError(w, err)
+		return
+	}
+	defer rows.Close()
+
+	var buf bytes.Buffer
+	zw := zip.NewWriter(&buf)
+	seen := map[string]int{} // de-duplicate filenames from same-titled docs
+	for rows.Next() {
+		var doc db.Document
+		if err := rows.Scan(
+			&doc.ID, &doc.UserID, &doc.Title, &doc.Content, &doc.ContentText,
+			&doc.Tone, &doc.WordCount, &doc.CreatedAt, &doc.UpdatedAt,
+		); err != nil {
+			serverError(w, err)
+			return
+		}
+		body, err := format.render(doc)
+		if err != nil {
+			serverError(w, err)
+			return
+		}
+		base := sanitizeFilename(doc.Title)
+		if base == "" {
+			base = "untitled"
+		}
+		key := base + "." + format.ext
+		name := key
+		if c := seen[key]; c > 0 {
+			name = fmt.Sprintf("%s (%d).%s", base, c, format.ext)
+		}
+		seen[key]++
+		f, err := zw.Create(name)
+		if err != nil {
+			serverError(w, err)
+			return
+		}
+		if _, err := f.Write(body); err != nil {
+			serverError(w, err)
+			return
+		}
+	}
+	if err := rows.Err(); err != nil {
+		serverError(w, err)
+		return
+	}
+	if err := zw.Close(); err != nil {
+		serverError(w, err)
+		return
+	}
+
+	filename := "petal-backup-" + time.Now().Format("2006-01-02") + ".zip"
+	w.Header().Set("Content-Type", "application/zip")
+	w.Header().Set("Content-Disposition", fmt.Sprintf("attachment; filename*=UTF-8''%s", urlEscapeFilename(filename)))
+	w.Header().Set("Content-Length", fmt.Sprintf("%d", buf.Len()))
+	_, _ = w.Write(buf.Bytes())
 }
 
 // exportFormat describes one downloadable format: how to render it and how to
@@ -92,7 +178,8 @@ type pmNode struct {
 }
 
 type pmMark struct {
-	Type string `json:"type"`
+	Type  string         `json:"type"`
+	Attrs map[string]any `json:"attrs"`
 }
 
 // parseDoc decodes Document.Content into a node tree. On empty or malformed JSON
@@ -131,6 +218,29 @@ func (n pmNode) hasMark(t string) bool {
 	return false
 }
 
+// attrStr returns a string-valued attribute (e.g. an image src), or "".
+func (n pmNode) attrStr(key string) string {
+	if n.Attrs == nil {
+		return ""
+	}
+	if s, ok := n.Attrs[key].(string); ok {
+		return s
+	}
+	return ""
+}
+
+// markAttr returns a string attribute from the named mark (e.g. a link href).
+func (n pmNode) markAttr(markType, key string) string {
+	for _, m := range n.Marks {
+		if m.Type == markType && m.Attrs != nil {
+			if s, ok := m.Attrs[key].(string); ok {
+				return s
+			}
+		}
+	}
+	return ""
+}
+
 func (n pmNode) level() int {
 	if n.Attrs == nil {
 		return 1
@@ -173,6 +283,11 @@ func mdBlock(n pmNode, depth int) string {
 		return "```\n" + textContent(n) + "\n```"
 	case "horizontalRule":
 		return "---"
+	case "image":
+		alt := n.attrStr("alt")
+		return fmt.Sprintf("![%s](%s)", alt, n.attrStr("src"))
+	case "table":
+		return mdTable(n)
 	case "bulletList", "orderedList":
 		var items []string
 		for i, item := range n.Content {
@@ -213,6 +328,55 @@ func mdInline(nodes []pmNode) string {
 	return b.String()
 }
 
+// mdTable renders a table node as a GitHub-flavored Markdown table. The first
+// row is used as the header (Markdown tables require one); a separator row is
+// inserted after it. Cell text is flattened to inline Markdown.
+func mdTable(n pmNode) string {
+	var rows [][]string
+	for _, row := range n.Content {
+		if row.Type != "tableRow" {
+			continue
+		}
+		var cells []string
+		for _, cell := range row.Content {
+			// A cell holds block children (usually one paragraph); flatten them.
+			var parts []string
+			for _, c := range cell.Content {
+				parts = append(parts, mdInline(c.Content))
+			}
+			// Escape pipes so cell content doesn't break the column layout.
+			cells = append(cells, strings.ReplaceAll(strings.TrimSpace(strings.Join(parts, " ")), "|", "\\|"))
+		}
+		rows = append(rows, cells)
+	}
+	if len(rows) == 0 {
+		return ""
+	}
+	cols := 0
+	for _, r := range rows {
+		if len(r) > cols {
+			cols = len(r)
+		}
+	}
+	pad := func(r []string) string {
+		for len(r) < cols {
+			r = append(r, "")
+		}
+		return "| " + strings.Join(r, " | ") + " |"
+	}
+	var lines []string
+	lines = append(lines, pad(rows[0]))
+	sep := make([]string, cols)
+	for i := range sep {
+		sep[i] = "---"
+	}
+	lines = append(lines, "| "+strings.Join(sep, " | ")+" |")
+	for _, r := range rows[1:] {
+		lines = append(lines, pad(r))
+	}
+	return strings.Join(lines, "\n")
+}
+
 func applyMdMarks(n pmNode) string {
 	t := n.Text
 	if n.hasMark("code") {
@@ -230,6 +394,9 @@ func applyMdMarks(n pmNode) string {
 	if n.hasMark("underline") {
 		t = "<u>" + t + "</u>"
 	}
+	if href := n.markAttr("link", "href"); href != "" {
+		t = "[" + t + "](" + href + ")"
+	}
 	return t
 }
 
@@ -347,6 +514,11 @@ func htmlBlock(n pmNode) string {
 		return "<pre><code>" + htmlEscape(textContent(n)) + "</code></pre>\n"
 	case "horizontalRule":
 		return "<hr>\n"
+	case "image":
+		alt := htmlEscape(n.attrStr("alt"))
+		return fmt.Sprintf("<p><img src=\"%s\" alt=\"%s\"></p>\n", htmlEscape(n.attrStr("src")), alt)
+	case "table":
+		return htmlTable(n)
 	case "bulletList", "orderedList":
 		tag := "ul"
 		if n.Type == "orderedList" {
@@ -376,6 +548,38 @@ func htmlBlock(n pmNode) string {
 	}
 }
 
+// htmlTable renders a table node as an HTML <table>. tableHeader cells become
+// <th>, tableCell cells become <td>; each cell's block children are flattened
+// to inline HTML.
+func htmlTable(n pmNode) string {
+	var b strings.Builder
+	b.WriteString("<table>\n")
+	for _, row := range n.Content {
+		if row.Type != "tableRow" {
+			continue
+		}
+		b.WriteString("<tr>")
+		for _, cell := range row.Content {
+			tag := "td"
+			if cell.Type == "tableHeader" {
+				tag = "th"
+			}
+			var inner strings.Builder
+			for _, c := range cell.Content {
+				if c.Type == "paragraph" {
+					inner.WriteString(htmlInline(c.Content))
+				} else {
+					inner.WriteString(htmlBlock(c))
+				}
+			}
+			b.WriteString("<" + tag + ">" + inner.String() + "</" + tag + ">")
+		}
+		b.WriteString("</tr>\n")
+	}
+	b.WriteString("</table>\n")
+	return b.String()
+}
+
 func htmlInline(nodes []pmNode) string {
 	var b strings.Builder
 	for _, n := range nodes {
@@ -408,6 +612,12 @@ func applyHTMLMarks(n pmNode) string {
 	if n.hasMark("strike") {
 		t = "<s>" + t + "</s>"
 	}
+	if n.hasMark("highlight") {
+		t = "<mark>" + t + "</mark>"
+	}
+	if href := n.markAttr("link", "href"); href != "" {
+		t = fmt.Sprintf("<a href=\"%s\">%s</a>", htmlEscape(href), t)
+	}
 	return t
 }
 
@@ -500,6 +710,16 @@ func docxBlock(n pmNode) string {
 		return b.String()
 	case "horizontalRule":
 		return `<w:p><w:pPr><w:pBdr><w:bottom w:val="single" w:sz="6" w:space="1" w:color="auto"/></w:pBdr></w:pPr></w:p>`
+	case "image":
+		// Image bytes aren't embedded (that needs media parts); leave a labeled
+		// placeholder so the reader knows an image belonged here.
+		label := n.attrStr("alt")
+		if label == "" {
+			label = "image"
+		}
+		return docxPara(pmNode{Content: []pmNode{{Type: "text", Text: "[" + label + "]", Marks: []pmMark{{Type: "italic"}}}}}, "")
+	case "table":
+		return docxTable(n)
 	case "bulletList", "orderedList":
 		var b strings.Builder
 		for i, item := range n.Content {
@@ -520,6 +740,48 @@ func docxBlock(n pmNode) string {
 	}
 }
 
+// docxTable renders a table node as a Word table (w:tbl) with single-line
+// borders. Header cells get a shaded background; every cell's block children are
+// rendered as paragraphs inside the cell.
+func docxTable(n pmNode) string {
+	var b strings.Builder
+	b.WriteString(`<w:tbl><w:tblPr><w:tblStyle w:val="TableGrid"/><w:tblW w:w="0" w:type="auto"/>` +
+		`<w:tblBorders>` +
+		`<w:top w:val="single" w:sz="4" w:space="0" w:color="auto"/>` +
+		`<w:left w:val="single" w:sz="4" w:space="0" w:color="auto"/>` +
+		`<w:bottom w:val="single" w:sz="4" w:space="0" w:color="auto"/>` +
+		`<w:right w:val="single" w:sz="4" w:space="0" w:color="auto"/>` +
+		`<w:insideH w:val="single" w:sz="4" w:space="0" w:color="auto"/>` +
+		`<w:insideV w:val="single" w:sz="4" w:space="0" w:color="auto"/>` +
+		`</w:tblBorders></w:tblPr>`)
+	for _, row := range n.Content {
+		if row.Type != "tableRow" {
+			continue
+		}
+		b.WriteString("<w:tr>")
+		for _, cell := range row.Content {
+			b.WriteString("<w:tc><w:tcPr>")
+			if cell.Type == "tableHeader" {
+				b.WriteString(`<w:shd w:val="clear" w:color="auto" w:fill="FFF0F5"/>`)
+			}
+			b.WriteString("</w:tcPr>")
+			// A cell must contain at least one paragraph to be valid.
+			if len(cell.Content) == 0 {
+				b.WriteString("<w:p/>")
+			}
+			for _, c := range cell.Content {
+				b.WriteString(docxBlock(c))
+			}
+			b.WriteString("</w:tc>")
+		}
+		b.WriteString("</w:tr>")
+	}
+	b.WriteString("</w:tbl>")
+	// Word needs a paragraph after a table; otherwise consecutive tables merge.
+	b.WriteString("<w:p/>")
+	return b.String()
+}
+
 // docxPara renders a block's inline children as a Word paragraph. An optional
 // style name (e.g. "Quote") and an optional literal text prefix (for list
 // markers) may be supplied.
@@ -571,6 +833,15 @@ func docxRun(n pmNode) string {
 	if n.hasMark("strike") {
 		props.WriteString("<w:strike/>")
 	}
+	if n.hasMark("highlight") {
+		// Word's text highlight only supports a fixed palette of named colors.
+		props.WriteString(`<w:highlight w:val="yellow"/>`)
+	}
+	if n.markAttr("link", "href") != "" {
+		// Without hyperlink relationships, style links as blue underlined text so
+		// they at least read as links (the URL itself is preserved in md/html).
+		props.WriteString(`<w:color w:val="2563EB"/><w:u w:val="single"/>`)
+	}
 	rpr := ""
 	if props.Len() > 0 {
 		rpr = "<w:rPr>" + props.String() + "</w:rPr>"

internal/docs/export_test.go

@@ -61,6 +61,45 @@ func TestExportMarkdown(t *testing.T) {
 	}
 }
 
+func TestExportAll(t *testing.T) {
+	srv := newTestServer(t)
+	// Two docs, one with a duplicate title to exercise filename de-duplication.
+	seedRichDoc(t, srv)
+	id2 := newDoc(t, srv)
+	if rec := do(t, srv, http.MethodPut, "/"+id2, `{"title":"日记 Diary","content":`+jsonString(richDocJSON)+`,"content_text":"x","word_count":1}`); rec.Code != http.StatusOK {
+		t.Fatalf("seed second doc: %d %s", rec.Code, rec.Body)
+	}
+
+	rec := do(t, srv, http.MethodGet, "/export-all?format=md", "")
+	if rec.Code != http.StatusOK {
+		t.Fatalf("export-all: code=%d body=%s", rec.Code, rec.Body)
+	}
+	if ct := rec.Header().Get("Content-Type"); ct != "application/zip" {
+		t.Fatalf("unexpected content-type: %q", ct)
+	}
+	zr, err := zip.NewReader(bytes.NewReader(rec.Body.Bytes()), int64(rec.Body.Len()))
+	if err != nil {
+		t.Fatalf("zip open: %v", err)
+	}
+	if len(zr.File) != 2 {
+		t.Fatalf("expected 2 files in backup, got %d", len(zr.File))
+	}
+	names := map[string]bool{}
+	for _, f := range zr.File {
+		names[f.Name] = true
+		rc, _ := f.Open()
+		b, _ := io.ReadAll(rc)
+		rc.Close()
+		if !strings.Contains(string(b), "你好") {
+			t.Fatalf("backup entry %q missing CJK body", f.Name)
+		}
+	}
+	// The duplicate title must have been disambiguated, not overwritten.
+	if !names["日记 Diary.md"] || !names["日记 Diary (1).md"] {
+		t.Fatalf("expected de-duplicated filenames, got %v", names)
+	}
+}
+
 func TestExportHTML(t *testing.T) {
 	srv := newTestServer(t)
 	id := seedRichDoc(t, srv)
@@ -126,6 +165,77 @@ func TestExportDocx(t *testing.T) {
 	}
 }
 
+// richDocJSON2 exercises the newer node/mark types: links, highlight, an image,
+// and a table (with a header row). Each export format should carry them through.
+const richDocJSON2 = `{"type":"doc","content":[` +
+	`{"type":"paragraph","content":[` +
+	`{"type":"text","marks":[{"type":"link","attrs":{"href":"https://petal.test"}}],"text":"site"},` +
+	`{"type":"text","text":" and "},` +
+	`{"type":"text","marks":[{"type":"highlight","attrs":{"color":"#FFF1A8"}}],"text":"lit"}` +
+	`]},` +
+	`{"type":"image","attrs":{"src":"/api/images/abc123.png","alt":"a cat"}},` +
+	`{"type":"table","content":[` +
+	`{"type":"tableRow","content":[` +
+	`{"type":"tableHeader","content":[{"type":"paragraph","content":[{"type":"text","text":"Name"}]}]},` +
+	`{"type":"tableHeader","content":[{"type":"paragraph","content":[{"type":"text","text":"年龄"}]}]}` +
+	`]},` +
+	`{"type":"tableRow","content":[` +
+	`{"type":"tableCell","content":[{"type":"paragraph","content":[{"type":"text","text":"Mei"}]}]},` +
+	`{"type":"tableCell","content":[{"type":"paragraph","content":[{"type":"text","text":"30"}]}]}` +
+	`]}` +
+	`]}` +
+	`]}`
+
+func seedRichDoc2(t *testing.T, srv http.Handler) string {
+	t.Helper()
+	id := newDoc(t, srv)
+	body := `{"title":"Rich2","content":` + jsonString(richDocJSON2) + `,"content_text":"site and lit\nName 年龄 Mei 30","word_count":6}`
+	if rec := do(t, srv, http.MethodPut, "/"+id, body); rec.Code != http.StatusOK {
+		t.Fatalf("seed rich doc 2: code=%d body=%s", rec.Code, rec.Body)
+	}
+	return id
+}
+
+func TestExportLinksImagesTables(t *testing.T) {
+	srv := newTestServer(t)
+	id := seedRichDoc2(t, srv)
+
+	md := do(t, srv, http.MethodGet, "/"+id+"/export?format=md", "").Body.String()
+	for _, want := range []string{"[site](https://petal.test)", "![a cat](/api/images/abc123.png)", "| Name | 年龄 |", "| --- | --- |", "| Mei | 30 |"} {
+		if !strings.Contains(md, want) {
+			t.Fatalf("markdown missing %q in:\n%s", want, md)
+		}
+	}
+
+	html := do(t, srv, http.MethodGet, "/"+id+"/export?format=html", "").Body.String()
+	for _, want := range []string{`<a href="https://petal.test">site</a>`, "<mark>lit</mark>", `<img src="/api/images/abc123.png" alt="a cat">`, "<th>Name</th>", "<td>Mei</td>"} {
+		if !strings.Contains(html, want) {
+			t.Fatalf("html missing %q in:\n%s", want, html)
+		}
+	}
+
+	docx := do(t, srv, http.MethodGet, "/"+id+"/export?format=docx", "")
+	raw := docx.Body.Bytes()
+	zr, err := zip.NewReader(bytes.NewReader(raw), int64(len(raw)))
+	if err != nil {
+		t.Fatalf("docx not a zip: %v", err)
+	}
+	var docXML string
+	for _, f := range zr.File {
+		if f.Name == "word/document.xml" {
+			rc, _ := f.Open()
+			b, _ := io.ReadAll(rc)
+			rc.Close()
+			docXML = string(b)
+		}
+	}
+	for _, want := range []string{"<w:tbl>", "Name", "年龄", "[a cat]", "<w:highlight"} {
+		if !strings.Contains(docXML, want) {
+			t.Fatalf("docx document.xml missing %q", want)
+		}
+	}
+}
+
 func TestExportUnsupportedFormat(t *testing.T) {
 	srv := newTestServer(t)
 	id := newDoc(t, srv)

internal/images/handler.go

@@ -0,0 +1,132 @@
+// Package images implements a tiny content-addressed image store: writers upload
+// images from the editor, they're saved to disk under the configured directory,
+// and served back by hashed filename. Content addressing means the same image
+// pasted twice is stored once, and URLs are stable and cacheable forever.
+package images
+
+import (
+	"crypto/sha256"
+	"encoding/hex"
+	"encoding/json"
+	"errors"
+	"io"
+	"net/http"
+	"os"
+	"path/filepath"
+	"strings"
+
+	"github.com/go-chi/chi/v5"
+)
+
+// maxUploadBytes caps a single image at 10 MiB — generous for a writing tool,
+// small enough to keep a careless paste from filling the disk.
+const maxUploadBytes = 10 << 20
+
+// extByContentType maps the image types we accept to a canonical extension. The
+// allowlist doubles as validation: anything not here is rejected.
+var extByContentType = map[string]string{
+	"image/png":     ".png",
+	"image/jpeg":    ".jpg",
+	"image/gif":     ".gif",
+	"image/webp":    ".webp",
+	"image/svg+xml": ".svg",
+}
+
+// Handler serves the upload + fetch endpoints, backed by a directory on disk.
+type Handler struct {
+	dir string
+}
+
+// New constructs a Handler, ensuring the storage directory exists.
+func New(dir string) (*Handler, error) {
+	if err := os.MkdirAll(dir, 0o755); err != nil {
+		return nil, err
+	}
+	return &Handler{dir: dir}, nil
+}
+
+// Routes mounts the image endpoints. Mount under "/images" so the full paths are
+// POST /api/images (upload) and GET /api/images/{name} (fetch).
+func (h *Handler) Routes() chi.Router {
+	r := chi.NewRouter()
+	r.Post("/", h.upload)
+	r.Get("/{name}", h.serve)
+	return r
+}
+
+// upload accepts a single multipart "image" field, sniffs and validates its
+// type, and writes it under a content hash so identical images dedupe. Responds
+// with the served URL the editor inserts.
+func (h *Handler) upload(w http.ResponseWriter, r *http.Request) {
+	r.Body = http.MaxBytesReader(w, r.Body, maxUploadBytes)
+	file, _, err := r.FormFile("image")
+	if err != nil {
+		http.Error(w, "expected an 'image' file field", http.StatusBadRequest)
+		return
+	}
+	defer file.Close()
+
+	data, err := io.ReadAll(file)
+	if err != nil {
+		http.Error(w, "could not read upload", http.StatusBadRequest)
+		return
+	}
+
+	// Trust a sniff over the client-declared type. SVG isn't reliably sniffable
+	// (DetectContentType returns text/plain or text/xml), so fall back to a
+	// lightweight tag check for it.
+	ct := http.DetectContentType(data)
+	ext, ok := extByContentType[ct]
+	if !ok {
+		if looksLikeSVG(data) {
+			ext, ok = ".svg", true
+		}
+	}
+	if !ok {
+		http.Error(w, "unsupported image type", http.StatusUnsupportedMediaType)
+		return
+	}
+
+	sum := sha256.Sum256(data)
+	name := hex.EncodeToString(sum[:])[:32] + ext
+	path := filepath.Join(h.dir, name)
+
+	// Skip the write if this exact content is already stored.
+	if _, statErr := os.Stat(path); errors.Is(statErr, os.ErrNotExist) {
+		if err := os.WriteFile(path, data, 0o644); err != nil {
+			http.Error(w, "could not store image", http.StatusInternalServerError)
+			return
+		}
+	}
+
+	w.Header().Set("Content-Type", "application/json")
+	_ = json.NewEncoder(w).Encode(map[string]string{"url": "/api/images/" + name})
+}
+
+// serve returns a stored image by its hashed filename. The filename is validated
+// to be a bare name (no path separators) so it can't escape the storage dir, and
+// served with a long-lived cache header since content-addressed URLs never change.
+func (h *Handler) serve(w http.ResponseWriter, r *http.Request) {
+	name := chi.URLParam(r, "name")
+	if name == "" || name != filepath.Base(name) || strings.ContainsAny(name, `/\`) {
+		http.NotFound(w, r)
+		return
+	}
+	path := filepath.Join(h.dir, filepath.Base(name))
+	if _, err := os.Stat(path); err != nil {
+		http.NotFound(w, r)
+		return
+	}
+	w.Header().Set("Cache-Control", "public, max-age=31536000, immutable")
+	http.ServeFile(w, r, path)
+}
+
+// looksLikeSVG does a cheap check for an <svg root tag near the start of the
+// file, since DetectContentType doesn't recognize SVG.
+func looksLikeSVG(data []byte) bool {
+	head := data
+	if len(head) > 512 {
+		head = head[:512]
+	}
+	return strings.Contains(strings.ToLower(string(head)), "<svg")
+}

internal/images/handler_test.go

@@ -0,0 +1,97 @@
+package images
+
+import (
+	"bytes"
+	"encoding/json"
+	"mime/multipart"
+	"net/http"
+	"net/http/httptest"
+	"strings"
+	"testing"
+)
+
+// a 1x1 transparent PNG.
+var pngBytes = []byte{
+	0x89, 0x50, 0x4e, 0x47, 0x0d, 0x0a, 0x1a, 0x0a, 0x00, 0x00, 0x00, 0x0d, 0x49, 0x48, 0x44, 0x52,
+	0x00, 0x00, 0x00, 0x01, 0x00, 0x00, 0x00, 0x01, 0x08, 0x06, 0x00, 0x00, 0x00, 0x1f, 0x15, 0xc4,
+	0x89, 0x00, 0x00, 0x00, 0x0d, 0x49, 0x44, 0x41, 0x54, 0x78, 0x9c, 0x62, 0x00, 0x01, 0x00, 0x00,
+	0x05, 0x00, 0x01, 0x0d, 0x0a, 0x2d, 0xb4, 0x00, 0x00, 0x00, 0x00, 0x49, 0x45, 0x4e, 0x44, 0xae,
+	0x42, 0x60, 0x82,
+}
+
+func uploadReq(t *testing.T, field string, data []byte) *http.Request {
+	t.Helper()
+	var buf bytes.Buffer
+	mw := multipart.NewWriter(&buf)
+	fw, err := mw.CreateFormFile(field, "x.png")
+	if err != nil {
+		t.Fatal(err)
+	}
+	fw.Write(data)
+	mw.Close()
+	req := httptest.NewRequest(http.MethodPost, "/", &buf)
+	req.Header.Set("Content-Type", mw.FormDataContentType())
+	return req
+}
+
+func TestUploadAndServe(t *testing.T) {
+	h, err := New(t.TempDir())
+	if err != nil {
+		t.Fatal(err)
+	}
+	r := h.Routes()
+
+	// Upload a PNG → expect a JSON url under /api/images/.
+	rec := httptest.NewRecorder()
+	r.ServeHTTP(rec, uploadReq(t, "image", pngBytes))
+	if rec.Code != http.StatusOK {
+		t.Fatalf("upload code=%d body=%s", rec.Code, rec.Body)
+	}
+	var resp struct{ URL string }
+	if err := json.Unmarshal(rec.Body.Bytes(), &resp); err != nil {
+		t.Fatal(err)
+	}
+	if !strings.HasPrefix(resp.URL, "/api/images/") || !strings.HasSuffix(resp.URL, ".png") {
+		t.Fatalf("unexpected url %q", resp.URL)
+	}
+
+	// The same content uploaded again dedupes to the same URL.
+	rec2 := httptest.NewRecorder()
+	r.ServeHTTP(rec2, uploadReq(t, "image", pngBytes))
+	var resp2 struct{ URL string }
+	json.Unmarshal(rec2.Body.Bytes(), &resp2)
+	if resp2.URL != resp.URL {
+		t.Fatalf("expected dedup to same url, got %q vs %q", resp2.URL, resp.URL)
+	}
+
+	// Fetch it back.
+	name := strings.TrimPrefix(resp.URL, "/api/images/")
+	rec3 := httptest.NewRecorder()
+	r.ServeHTTP(rec3, httptest.NewRequest(http.MethodGet, "/"+name, nil))
+	if rec3.Code != http.StatusOK {
+		t.Fatalf("serve code=%d", rec3.Code)
+	}
+	if !bytes.Equal(rec3.Body.Bytes(), pngBytes) {
+		t.Fatal("served bytes differ from uploaded")
+	}
+}
+
+func TestUploadRejectsNonImage(t *testing.T) {
+	h, _ := New(t.TempDir())
+	r := h.Routes()
+	rec := httptest.NewRecorder()
+	r.ServeHTTP(rec, uploadReq(t, "image", []byte("this is plainly not an image at all")))
+	if rec.Code != http.StatusUnsupportedMediaType {
+		t.Fatalf("expected 415, got %d", rec.Code)
+	}
+}
+
+func TestServeMissing(t *testing.T) {
+	h, _ := New(t.TempDir())
+	r := h.Routes()
+	rec := httptest.NewRecorder()
+	r.ServeHTTP(rec, httptest.NewRequest(http.MethodGet, "/deadbeef.png", nil))
+	if rec.Code != http.StatusNotFound {
+		t.Fatalf("expected 404, got %d", rec.Code)
+	}
+}

internal/lexicon/data.go

@@ -27,3 +27,12 @@ var synonymsGz []byte
 //
 //go:embed data/gloss.json.gz
 var glossGz []byte
+
+// phoneticGz is the gzipped English-phonetic map: word → IPA, lowercase keys.
+// Built from ECDICT's phonetic column (scripts/build_phonetic.py). Shown beside
+// the read-aloud button so the writer — a Mandarin speaker learning English —
+// can see how to say the word. Ships with a small common-word seed; a full build
+// broadens coverage.
+//
+//go:embed data/phonetic.json.gz
+var phoneticGz []byte

internal/lexicon/lexicon.go

@@ -26,6 +26,7 @@ type Meaning struct {
 type Result struct {
 	Word        string    `json:"word"`
 	Gloss       string    `json:"gloss"`
+	Phonetic    string    `json:"phonetic"` // IPA for the English word; "" when absent
 	Definitions []Meaning `json:"definitions"`
 	Synonyms    []string  `json:"synonyms"`
 }
@@ -56,6 +57,7 @@ type Lexicon struct {
 	defs     map[string][][]string // word → [[pos, def, example], …]
 	synonyms map[string][]string   // word → [synonym, …]
 	gloss    map[string]string     // word → Chinese gloss
+	phonetic map[string]string     // word → IPA
 }
 
 // New returns a Lexicon. The datasets aren't read until the first Lookup.
@@ -75,6 +77,10 @@ func (l *Lexicon) load() {
 			l.loadErr = fmt.Errorf("load gloss: %w", err)
 			return
 		}
+		if err := gunzipJSON(phoneticGz, &l.phonetic); err != nil {
+			l.loadErr = fmt.Errorf("load phonetic: %w", err)
+			return
+		}
 	})
 }
 
@@ -95,6 +101,9 @@ func (l *Lexicon) Lookup(word string) (Result, error) {
 	}
 
 	res.Gloss = lookupGloss(l.gloss, norm)
+	// Phonetic is a word→string map like the gloss, so the same de-inflecting
+	// candidate walk resolves "running" → "run", etc.
+	res.Phonetic = lookupGloss(l.phonetic, norm)
 
 	if raw := lookupDefs(l.defs, norm); raw != nil {
 		for _, m := range raw {

internal/suggestions/handlers.go

@@ -154,6 +154,14 @@ var (
 // replacePending swaps a document's pending suggestions within one family for a
 // fresh batch in a single transaction. Accepted/rejected suggestions and the
 // other family's pending rows are left untouched.
+//
+// Suggestions the user already accepted or dismissed are suppressed from the
+// fresh batch: the model has no memory between passes, so without this it would
+// re-propose the identical edit on the very next checkpoint — re-nagging a
+// sentence the user already resolved. This matters most when an accept silently
+// no-ops (the `original` text couldn't be anchored in the editor, so the doc
+// never changed): the sentence is unaltered, yet the user shouldn't see the same
+// card again.
 func (h *Handler) replacePending(docID, contentText string, raw []llm.RawSuggestion, scope pendingScope) error {
 	tx, err := h.DB.Begin()
 	if err != nil {
@@ -168,7 +176,15 @@ func (h *Handler) replacePending(docID, contentText string, raw []llm.RawSuggest
 		return err
 	}
 
+	actioned, err := actionedKeys(tx, docID)
+	if err != nil {
+		return err
+	}
+
 	for _, s := range raw {
+		if _, seen := actioned[suggestionKey(s.Original, s.Replacement)]; seen {
+			continue
+		}
 		typ := scope.forceType
 		if typ == "" {
 			typ = normalizeType(s.Type)
@@ -186,6 +202,39 @@ func (h *Handler) replacePending(docID, contentText string, raw []llm.RawSuggest
 	return tx.Commit()
 }
 
+// actionedKeys returns the set of original→replacement keys the user has already
+// accepted or rejected for this document, so a fresh checkpoint can skip
+// re-proposing them. Keyed on the trimmed original+replacement pair, matching the
+// normalization ParseCheckpoint applies, so a genuinely different edit on the same
+// sentence is not suppressed.
+func actionedKeys(tx *sql.Tx, docID string) (map[string]struct{}, error) {
+	rows, err := tx.Query(
+		`SELECT original, replacement FROM suggestions
+		  WHERE doc_id = ? AND status IN (?, ?)`,
+		docID, db.SuggestionStatusAccepted, db.SuggestionStatusRejected,
+	)
+	if err != nil {
+		return nil, err
+	}
+	defer rows.Close()
+
+	keys := make(map[string]struct{})
+	for rows.Next() {
+		var original, replacement string
+		if err := rows.Scan(&original, &replacement); err != nil {
+			return nil, err
+		}
+		keys[suggestionKey(original, replacement)] = struct{}{}
+	}
+	return keys, rows.Err()
+}
+
+// suggestionKey is the dedup identity for a suggestion: its trimmed original and
+// replacement text. Two suggestions with the same key are "the same edit."
+func suggestionKey(original, replacement string) string {
+	return strings.TrimSpace(original) + "\x00" + strings.TrimSpace(replacement)
+}
+
 // listForDoc returns the document's current pending suggestions (used when the
 // editor loads a document, before any new checkpoint fires).
 func (h *Handler) listForDoc(w http.ResponseWriter, r *http.Request) {

internal/suggestions/handlers_test.go

@@ -127,6 +127,44 @@ func TestCheckpointFlow(t *testing.T) {
 	}
 }
 
+// TestResolvedSuggestionsNotReproposed proves a checkpoint won't re-nag a
+// sentence the user already accepted or dismissed: the model returns the same
+// raw batch on the next pass (it has no memory), but the resolved edits are
+// suppressed. This is the "accept it, then it nags again moments later" bug.
+func TestResolvedSuggestionsNotReproposed(t *testing.T) {
+	client := &stubClient{response: `{"suggestions":[
+		{"original":"I has","replacement":"I have","explanation":"subject-verb agreement","type":"grammar"},
+		{"original":"two apple","replacement":"two apples","explanation":"plural noun","type":"grammar"}
+	]}`}
+	srv, docID, h := newTestServer(t, client)
+	// Zero the checkpoint floor so the second pass runs instead of being throttled.
+	h.Limit = llm.NewRateLimiter(0)
+
+	rec := do(t, srv, http.MethodPost, "/docs/"+docID+"/check", "")
+	var got []db.Suggestion
+	_ = json.Unmarshal(rec.Body.Bytes(), &got)
+	if len(got) != 2 {
+		t.Fatalf("first pass: want 2, got %d", len(got))
+	}
+
+	// Accept one, dismiss the other.
+	do(t, srv, http.MethodPost, "/suggestions/"+got[0].ID+"/accept", "")
+	do(t, srv, http.MethodPost, "/suggestions/"+got[1].ID+"/dismiss", "")
+
+	// Second checkpoint returns the identical raw batch — both must be suppressed.
+	rec = do(t, srv, http.MethodPost, "/docs/"+docID+"/check", "")
+	if rec.Code != http.StatusOK {
+		t.Fatalf("second check: code=%d body=%s", rec.Code, rec.Body)
+	}
+	var again []db.Suggestion
+	if err := json.Unmarshal(rec.Body.Bytes(), &again); err != nil {
+		t.Fatalf("decode: %v", err)
+	}
+	if len(again) != 0 {
+		t.Fatalf("resolved suggestions should not be re-proposed, got %d: %+v", len(again), again)
+	}
+}
+
 // TestVoicePassCoexists proves the voice pass and the grammar checkpoint own
 // disjoint pending families: running one never wipes the other's flags, and
 // each endpoint returns the unified pending set.

internal/tts/handler.go

@@ -0,0 +1,264 @@
+// Package tts implements read-aloud: it proxies short passages to a local Piper
+// HTTP server (a fast, offline neural TTS), optionally transcodes Piper's WAV to
+// mp3/opus with ffmpeg, and serves the audio back to the editor. Synthesized
+// clips are content-addressed on disk so tapping the same word twice is instant
+// and never re-synthesizes. The feature is entirely optional: when TTS_ENDPOINT
+// is unset the route isn't mounted and the frontend falls back to the browser's
+// Web Speech API.
+package tts
+
+import (
+	"bytes"
+	"context"
+	"crypto/sha256"
+	"encoding/hex"
+	"encoding/json"
+	"fmt"
+	"io"
+	"net/http"
+	"os"
+	"os/exec"
+	"path/filepath"
+	"strings"
+	"time"
+	"unicode/utf8"
+
+	"github.com/go-chi/chi/v5"
+
+	"gitea.parodia.dev/drwily/petal/internal/config"
+)
+
+// maxTextBytes bounds a single synthesis request. Read-aloud is for a word or a
+// sentence or two, not whole documents; this keeps one tap from pinning Piper.
+const maxTextBytes = 4000
+
+// lengthScale slows Piper slightly below natural pace — a learner following along
+// with the words, mirroring the old utterance.rate = 0.95. Higher = slower.
+const lengthScale = 1.1
+
+// audioFormat describes one output encoding: the cache-file extension, the
+// response Content-Type, and the ffmpeg args that turn Piper's WAV (on stdin)
+// into this format (on stdout). A nil ffmpegArgs means "serve the WAV as-is".
+type audioFormat struct {
+	ext         string
+	contentType string
+	ffmpegArgs  []string
+}
+
+// formats maps the configured TTS_AUDIO_FORMAT to its encoding. mp3 is the safe
+// default (every browser plays it); opus is ~half the size for speech but has
+// patchier Safari support; wav skips ffmpeg entirely.
+var formats = map[string]audioFormat{
+	"wav": {ext: ".wav", contentType: "audio/wav"},
+	"mp3": {ext: ".mp3", contentType: "audio/mpeg",
+		ffmpegArgs: []string{"-f", "mp3", "-c:a", "libmp3lame", "-b:a", "64k", "-ac", "1"}},
+	"opus": {ext: ".opus", contentType: "audio/ogg",
+		ffmpegArgs: []string{"-f", "ogg", "-c:a", "libopus", "-b:a", "32k", "-ac", "1"}},
+}
+
+// route is the Piper instance and voice id serving one language. Each Piper
+// HTTP server loads exactly one model, so distinct languages mean distinct
+// endpoints (e.g. English on :5005, Chinese on :5006).
+type route struct {
+	endpoint string // Piper HTTP base URL (no trailing slash)
+	voice    string // Piper voice id (sent in the request; also part of the cache key)
+}
+
+// Handler proxies synthesis to Piper and caches the result on disk.
+type Handler struct {
+	routes   map[string]route // base language (e.g. "en", "zh") -> Piper instance
+	cacheDir string
+	format   audioFormat
+	client   *http.Client
+}
+
+// New builds a Handler from config. It returns (nil, false) when TTS_ENDPOINT is
+// unset, signalling main to skip mounting the route. An unknown TTS_AUDIO_FORMAT
+// falls back to mp3 rather than failing the whole server.
+func New(cfg *config.Config) (*Handler, bool) {
+	if strings.TrimSpace(cfg.TTSEndpoint) == "" {
+		return nil, false
+	}
+
+	format, ok := formats[strings.ToLower(strings.TrimSpace(cfg.TTSFormat))]
+	if !ok {
+		format = formats["mp3"]
+	}
+
+	// Map by base language so en-US, en-GB, etc. all resolve to the English
+	// instance (the client sends BCP-47 tags like the old Web Speech path did).
+	// A language is only routable when both its endpoint and voice are set;
+	// otherwise the client falls back to Web Speech for that language.
+	routes := map[string]route{}
+	if cfg.TTSVoiceEN != "" {
+		routes["en"] = route{strings.TrimRight(cfg.TTSEndpoint, "/"), cfg.TTSVoiceEN}
+	}
+	if cfg.TTSEndpointZH != "" && cfg.TTSVoiceZH != "" {
+		routes["zh"] = route{strings.TrimRight(cfg.TTSEndpointZH, "/"), cfg.TTSVoiceZH}
+	}
+
+	if err := os.MkdirAll(cfg.TTSCacheDir, 0o755); err != nil {
+		// A missing cache dir isn't fatal — synthesis still works, it just won't
+		// cache. Disable the feature only on endpoint absence, not this.
+		fmt.Fprintf(os.Stderr, "tts: cache dir %q: %v\n", cfg.TTSCacheDir, err)
+	}
+
+	return &Handler{
+		routes:   routes,
+		cacheDir: cfg.TTSCacheDir,
+		format:   format,
+		client:   &http.Client{Timeout: cfg.TTSTimeout},
+	}, true
+}
+
+// Routes mounts the synthesis endpoint. Mount under "/tts" so the full path is
+// POST /api/tts.
+func (h *Handler) Routes() chi.Router {
+	r := chi.NewRouter()
+	r.Post("/", h.synth)
+	return r
+}
+
+// synthRequest is the body the editor posts: a passage and the BCP-47 language
+// tag it's written in (e.g. "en-US", "zh-CN").
+type synthRequest struct {
+	Text string `json:"text"`
+	Lang string `json:"lang"`
+}
+
+// synth resolves a voice for the requested language, returns cached audio when
+// present, and otherwise asks Piper to synthesize, transcodes if configured,
+// caches, and serves. An unconfigured language yields 404 so the client can fall
+// back to Web Speech without treating it as an error.
+func (h *Handler) synth(w http.ResponseWriter, r *http.Request) {
+	var req synthRequest
+	// Cap the raw body generously above maxTextBytes to leave room for JSON
+	// framing and the lang field; the text itself is truncated after decoding.
+	if err := json.NewDecoder(io.LimitReader(r.Body, maxTextBytes+1024)).Decode(&req); err != nil {
+		http.Error(w, "invalid request body", http.StatusBadRequest)
+		return
+	}
+	text := strings.TrimSpace(req.Text)
+	if text == "" {
+		http.Error(w, "empty text", http.StatusBadRequest)
+		return
+	}
+	if len(text) > maxTextBytes {
+		// Trim back to the last valid rune boundary so we never hand Piper a
+		// half-encoded multibyte character (common with Chinese, 3 bytes/char).
+		text = text[:maxTextBytes]
+		for len(text) > 0 && !utf8.ValidString(text) {
+			text = text[:len(text)-1]
+		}
+	}
+
+	rt, ok := h.routes[baseLang(req.Lang)]
+	if !ok {
+		// No voice for this language — let the client fall back to Web Speech.
+		http.Error(w, "no voice for language", http.StatusNotFound)
+		return
+	}
+
+	// Content-addressed: identical (voice, text) → identical clip. The format
+	// extension keeps encodings from colliding in the same dir.
+	sum := sha256.Sum256([]byte(rt.voice + "\n" + text))
+	name := hex.EncodeToString(sum[:])[:32] + h.format.ext
+	path := filepath.Join(h.cacheDir, name)
+
+	if _, err := os.Stat(path); err == nil {
+		h.serve(w, r, path)
+		return
+	}
+
+	audio, err := h.synthesize(r.Context(), rt, text)
+	if err != nil {
+		http.Error(w, "synthesis failed", http.StatusBadGateway)
+		fmt.Fprintf(os.Stderr, "tts: synthesize: %v\n", err)
+		return
+	}
+
+	// Best-effort cache write via a temp file + rename so a concurrent reader
+	// never sees a half-written clip. A failed write just means no caching.
+	if h.cacheDir != "" {
+		tmp := path + ".tmp"
+		if err := os.WriteFile(tmp, audio, 0o644); err == nil {
+			_ = os.Rename(tmp, path)
+		}
+	}
+
+	w.Header().Set("Content-Type", h.format.contentType)
+	w.Header().Set("Cache-Control", "public, max-age=31536000, immutable")
+	_, _ = w.Write(audio)
+}
+
+// serve streams a cached clip with a long-lived immutable cache header (the URL
+// is content-addressed, so the bytes never change for a given request).
+func (h *Handler) serve(w http.ResponseWriter, r *http.Request, path string) {
+	w.Header().Set("Content-Type", h.format.contentType)
+	w.Header().Set("Cache-Control", "public, max-age=31536000, immutable")
+	http.ServeFile(w, r, path)
+}
+
+// synthesize POSTs to the route's Piper instance, then transcodes the returned
+// WAV when the configured format calls for it.
+func (h *Handler) synthesize(ctx context.Context, rt route, text string) ([]byte, error) {
+	body, _ := json.Marshal(map[string]any{
+		"text":         text,
+		"voice":        rt.voice,
+		"length_scale": lengthScale,
+	})
+	httpReq, err := http.NewRequestWithContext(ctx, http.MethodPost, rt.endpoint+"/", bytes.NewReader(body))
+	if err != nil {
+		return nil, err
+	}
+	httpReq.Header.Set("Content-Type", "application/json")
+
+	resp, err := h.client.Do(httpReq)
+	if err != nil {
+		return nil, err
+	}
+	defer resp.Body.Close()
+	if resp.StatusCode != http.StatusOK {
+		snippet, _ := io.ReadAll(io.LimitReader(resp.Body, 256))
+		return nil, fmt.Errorf("piper %d: %s", resp.StatusCode, strings.TrimSpace(string(snippet)))
+	}
+	wav, err := io.ReadAll(resp.Body)
+	if err != nil {
+		return nil, err
+	}
+
+	if h.format.ffmpegArgs == nil {
+		return wav, nil
+	}
+	return transcode(ctx, wav, h.format.ffmpegArgs)
+}
+
+// transcode pipes WAV bytes through ffmpeg (stdin → stdout) into the target
+// encoding. ffmpeg is assumed on PATH; an error here surfaces as a 502.
+func transcode(ctx context.Context, wav []byte, args []string) ([]byte, error) {
+	// Guard ffmpeg against a hang independent of the HTTP client timeout.
+	ctx, cancel := context.WithTimeout(ctx, 30*time.Second)
+	defer cancel()
+
+	full := append([]string{"-hide_banner", "-loglevel", "error", "-i", "pipe:0"}, args...)
+	full = append(full, "pipe:1")
+	cmd := exec.CommandContext(ctx, "ffmpeg", full...)
+	cmd.Stdin = bytes.NewReader(wav)
+	var out, errBuf bytes.Buffer
+	cmd.Stdout = &out
+	cmd.Stderr = &errBuf
+	if err := cmd.Run(); err != nil {
+		return nil, fmt.Errorf("ffmpeg: %v: %s", err, strings.TrimSpace(errBuf.String()))
+	}
+	return out.Bytes(), nil
+}
+
+// baseLang reduces a BCP-47 tag to its primary subtag, lowercased: "en-US" → "en",
+// "zh-CN" → "zh", "" → "". Mirrors the client's pickVoice base-language matching.
+func baseLang(tag string) string {
+	tag = strings.ToLower(strings.TrimSpace(tag))
+	if i := strings.IndexAny(tag, "-_"); i >= 0 {
+		return tag[:i]
+	}
+	return tag
+}

internal/tts/handler_test.go

@@ -0,0 +1,170 @@
+package tts
+
+import (
+	"bytes"
+	"encoding/json"
+	"net/http"
+	"net/http/httptest"
+	"strings"
+	"sync/atomic"
+	"testing"
+)
+
+// newStubPiper returns a fake Piper server that echoes a fixed WAV body and
+// records how many times it was called and the last request payload.
+func newStubPiper(t *testing.T, body []byte) (*httptest.Server, *int32, *synthEcho) {
+	t.Helper()
+	var calls int32
+	last := &synthEcho{}
+	srv := httptest.NewServer(http.HandlerFunc(func(w http.ResponseWriter, r *http.Request) {
+		atomic.AddInt32(&calls, 1)
+		var req map[string]any
+		_ = json.NewDecoder(r.Body).Decode(&req)
+		last.voice, _ = req["voice"].(string)
+		last.text, _ = req["text"].(string)
+		w.Header().Set("Content-Type", "audio/wav")
+		_, _ = w.Write(body)
+	}))
+	t.Cleanup(srv.Close)
+	return srv, &calls, last
+}
+
+type synthEcho struct{ voice, text string }
+
+// newHandler builds a wav-format handler (no ffmpeg) pointed at a stub server.
+func newHandler(t *testing.T, endpoint string) *Handler {
+	t.Helper()
+	return &Handler{
+		routes:   map[string]route{"en": {strings.TrimRight(endpoint, "/"), "en_US-amy-medium"}},
+		cacheDir: t.TempDir(),
+		format:   formats["wav"],
+		client:   http.DefaultClient,
+	}
+}
+
+func post(t *testing.T, h *Handler, text, lang string) *httptest.ResponseRecorder {
+	t.Helper()
+	b, _ := json.Marshal(synthRequest{Text: text, Lang: lang})
+	req := httptest.NewRequest(http.MethodPost, "/", bytes.NewReader(b))
+	rr := httptest.NewRecorder()
+	h.synth(rr, req)
+	return rr
+}
+
+func TestSynthSuccessAndVoiceSelection(t *testing.T) {
+	wav := []byte("RIFF....fake-wav")
+	srv, calls, last := newStubPiper(t, wav)
+	h := newHandler(t, srv.URL)
+
+	rr := post(t, h, "hello there", "en-US")
+	if rr.Code != http.StatusOK {
+		t.Fatalf("status = %d, want 200", rr.Code)
+	}
+	if got := rr.Body.Bytes(); !bytes.Equal(got, wav) {
+		t.Fatalf("body = %q, want the piper wav", got)
+	}
+	if ct := rr.Header().Get("Content-Type"); ct != "audio/wav" {
+		t.Fatalf("content-type = %q, want audio/wav", ct)
+	}
+	if last.voice != "en_US-amy-medium" {
+		t.Fatalf("piper voice = %q, want en_US-amy-medium", last.voice)
+	}
+	if *calls != 1 {
+		t.Fatalf("piper calls = %d, want 1", *calls)
+	}
+}
+
+func TestRoutesByLanguageToSeparateInstances(t *testing.T) {
+	enSrv, enCalls, _ := newStubPiper(t, []byte("EN-wav"))
+	zhSrv, zhCalls, zhLast := newStubPiper(t, []byte("ZH-wav"))
+	h := &Handler{
+		routes: map[string]route{
+			"en": {strings.TrimRight(enSrv.URL, "/"), "en_US-amy-medium"},
+			"zh": {strings.TrimRight(zhSrv.URL, "/"), "zh_CN-huayan-medium"},
+		},
+		cacheDir: t.TempDir(),
+		format:   formats["wav"],
+		client:   http.DefaultClient,
+	}
+
+	if rr := post(t, h, "你好世界", "zh-CN"); rr.Code != http.StatusOK {
+		t.Fatalf("zh status = %d, want 200", rr.Code)
+	}
+	if *zhCalls != 1 || *enCalls != 0 {
+		t.Fatalf("calls en=%d zh=%d, want en=0 zh=1 (zh routed to zh instance)", *enCalls, *zhCalls)
+	}
+	if zhLast.voice != "zh_CN-huayan-medium" {
+		t.Fatalf("zh voice = %q, want zh_CN-huayan-medium", zhLast.voice)
+	}
+}
+
+func TestUnknownLanguageReturns404(t *testing.T) {
+	srv, calls, _ := newStubPiper(t, []byte("x"))
+	h := newHandler(t, srv.URL)
+
+	rr := post(t, h, "你好", "zh-CN") // only "en" is configured
+	if rr.Code != http.StatusNotFound {
+		t.Fatalf("status = %d, want 404", rr.Code)
+	}
+	if *calls != 0 {
+		t.Fatalf("piper calls = %d, want 0 (should not synthesize)", *calls)
+	}
+}
+
+func TestCacheHitSkipsPiper(t *testing.T) {
+	srv, calls, _ := newStubPiper(t, []byte("RIFF....fake-wav"))
+	h := newHandler(t, srv.URL)
+
+	if rr := post(t, h, "same words", "en-US"); rr.Code != http.StatusOK {
+		t.Fatalf("first status = %d, want 200", rr.Code)
+	}
+	if rr := post(t, h, "same words", "en-US"); rr.Code != http.StatusOK {
+		t.Fatalf("second status = %d, want 200", rr.Code)
+	}
+	if *calls != 1 {
+		t.Fatalf("piper calls = %d, want 1 (second served from cache)", *calls)
+	}
+}
+
+func TestEmptyTextReturns400(t *testing.T) {
+	srv, _, _ := newStubPiper(t, []byte("x"))
+	h := newHandler(t, srv.URL)
+
+	if rr := post(t, h, "   ", "en-US"); rr.Code != http.StatusBadRequest {
+		t.Fatalf("status = %d, want 400", rr.Code)
+	}
+}
+
+func TestTextIsCapped(t *testing.T) {
+	srv, _, last := newStubPiper(t, []byte("x"))
+	h := newHandler(t, srv.URL)
+
+	long := strings.Repeat("a", maxTextBytes+500)
+	if rr := post(t, h, long, "en-US"); rr.Code != http.StatusOK {
+		t.Fatalf("status = %d, want 200", rr.Code)
+	}
+	if len(last.text) != maxTextBytes {
+		t.Fatalf("piper received %d chars, want capped to %d", len(last.text), maxTextBytes)
+	}
+}
+
+func TestBaseLang(t *testing.T) {
+	cases := map[string]string{"en-US": "en", "EN_gb": "en", "zh-CN": "zh", "en": "en", "": ""}
+	for in, want := range cases {
+		if got := baseLang(in); got != want {
+			t.Errorf("baseLang(%q) = %q, want %q", in, got, want)
+		}
+	}
+}
+
+// ensure the stub's body is fully consumable (guards against the LimitReader cap
+// accidentally truncating a normal request body in synth()).
+func TestRequestBodyNotTruncated(t *testing.T) {
+	srv, _, last := newStubPiper(t, []byte("x"))
+	h := newHandler(t, srv.URL)
+	text := strings.Repeat("word ", 200) // ~1000 bytes, under the cap
+	post(t, h, text, "en-US")
+	if last.text != strings.TrimSpace(text) {
+		t.Fatalf("piper text length = %d, want %d", len(last.text), len(strings.TrimSpace(text)))
+	}
+}

scripts/build_phonetic.py

@@ -0,0 +1,101 @@
+#!/usr/bin/env python3
+"""Build the embedded English-phonetic dataset.
+
+Petal's writer is a native Mandarin speaker learning English, so the useful
+pronunciation aid is *how to say the English word* — shown in the word popover
+beside the 🔊 read-aloud button. (Pinyin would annotate Chinese she already
+reads fluently, so it isn't built.) The phonetic spelling comes from the same
+ECDICT source as the Chinese gloss, which already carries a `phonetic` column.
+
+Two modes:
+
+  # Full build from ECDICT (same CSV used by build_gloss.py):
+  curl -sL https://raw.githubusercontent.com/skywind3000/ECDICT/master/ecdict.csv -o ecdict.csv
+  python3 scripts/build_phonetic.py ecdict.csv internal/lexicon/data/phonetic.json.gz
+
+  # Write just the built-in common-word seed (no CSV needed — what ships in-repo
+  # so the feature works out of the box until a full build is run):
+  python3 scripts/build_phonetic.py --seed internal/lexicon/data/phonetic.json.gz
+"""
+import csv
+import gzip
+import json
+import re
+import sys
+
+# Same frequency gate as the gloss build, so coverage matches the words an ESL
+# writer actually meets and the asset stays small.
+MAX_RANK = 50000
+
+# Single English word (letters with internal apostrophe/hyphen).
+WORD_RE = re.compile(r"^[a-z][a-z'\-]*[a-z]$|^[a-z]$")
+
+# A small, hand-checked seed of common words → IPA. This ships in the repo so the
+# phonetic line is live immediately; a full ECDICT build overwrites it with broad
+# coverage. Kept deliberately short and correct rather than large and shaky.
+SEED = {
+    "hello": "həˈloʊ", "world": "wɜːrld", "people": "ˈpiːpl", "water": "ˈwɔːtər",
+    "river": "ˈrɪvər", "house": "haʊs", "school": "skuːl", "friend": "frɛnd",
+    "family": "ˈfæməli", "mother": "ˈmʌðər", "father": "ˈfɑːðər", "woman": "ˈwʊmən",
+    "language": "ˈlæŋɡwɪdʒ", "english": "ˈɪŋɡlɪʃ", "write": "raɪt", "writing": "ˈraɪtɪŋ",
+    "read": "riːd", "word": "wɜːrd", "sentence": "ˈsɛntəns", "beautiful": "ˈbjuːtɪfl",
+    "happy": "ˈhæpi", "love": "lʌv", "thought": "θɔːt", "through": "θruː",
+    "though": "ðoʊ", "enough": "ɪˈnʌf", "question": "ˈkwɛstʃən", "answer": "ˈænsər",
+    "because": "bɪˈkɔːz", "people's": "ˈpiːplz", "important": "ɪmˈpɔːrtnt",
+    "different": "ˈdɪfərənt", "remember": "rɪˈmɛmbər", "together": "təˈɡɛðər",
+    "morning": "ˈmɔːrnɪŋ", "tonight": "təˈnaɪt", "yesterday": "ˈjɛstərdeɪ",
+    "tomorrow": "təˈmɒroʊ", "weather": "ˈwɛðər", "color": "ˈkʌlər", "colour": "ˈkʌlər",
+    "favourite": "ˈfeɪvərɪt", "favorite": "ˈfeɪvərɪt", "comfortable": "ˈkʌmftəbl",
+    "vegetable": "ˈvɛdʒtəbl", "interesting": "ˈɪntrəstɪŋ", "necessary": "ˈnɛsəsɛri",
+    "february": "ˈfɛbruɛri", "wednesday": "ˈwɛnzdeɪ",
+    "island": "ˈaɪlənd", "knife": "naɪf", "know": "noʊ", "knee": "niː",
+    "hour": "ˈaʊər", "honest": "ˈɒnɪst", "listen": "ˈlɪsn", "castle": "ˈkæsl",
+    "thank": "θæŋk", "think": "θɪŋk", "thing": "θɪŋ", "three": "θriː",
+    "voice": "vɔɪs", "choice": "tʃɔɪs", "nature": "ˈneɪtʃər", "picture": "ˈpɪktʃər",
+    "future": "ˈfjuːtʃər", "culture": "ˈkʌltʃər", "measure": "ˈmɛʒər",
+    "usually": "ˈjuːʒuəli", "decision": "dɪˈsɪʒn", "delicious": "dɪˈlɪʃəs",
+}
+
+
+def clean_phonetic(p: str) -> str:
+    """Normalize an ECDICT phonetic field: trim, strip wrapping slashes/brackets."""
+    p = p.strip().strip("/[]").strip()
+    return p
+
+
+def build_from_csv(src: str) -> dict:
+    out = {}
+    with open(src, newline="", encoding="utf-8") as fh:
+        for row in csv.DictReader(fh):
+            word = (row.get("word") or "").strip().lower()
+            phon = clean_phonetic(row.get("phonetic") or "")
+            if not word or not phon or not WORD_RE.match(word):
+                continue
+            try:
+                rank = int(row.get("frq") or 0) or int(row.get("bnc") or 0)
+            except ValueError:
+                rank = 0
+            if rank <= 0 or rank > MAX_RANK:
+                continue
+            out[word] = phon
+    # Make sure the curated seed is always present (and authoritative).
+    out.update(SEED)
+    return out
+
+
+def main() -> None:
+    args = sys.argv[1:]
+    if not args:
+        sys.exit(__doc__)
+    if args[0] == "--seed":
+        data, dest = dict(SEED), args[1]
+    else:
+        src, dest = args[0], args[1]
+        data = build_from_csv(src)
+    with gzip.open(dest, "wt", encoding="utf-8") as fh:
+        json.dump(data, fh, ensure_ascii=False, separators=(",", ":"))
+    print(f"wrote {len(data)} phonetic entries to {dest}")
+
+
+if __name__ == "__main__":
+    main()

scripts/gen_sounds.py

@@ -0,0 +1,165 @@
+#!/usr/bin/env python3
+"""Generate Petal's cute UI sound palette as small WAV assets.
+
+These are warm, rounded little sounds — bubble pops, water drops, soft bells —
+played when suggestions and companion bubbles appear. We synthesize them here
+(rather than ship someone else's clips) so they match the app's gentle aesthetic
+exactly and stay tiny. Output lands in web/src/assets/sounds/ where Vite bundles
+and fingerprints them into dist (and so into the embedded Go binary).
+
+Run:  python3 scripts/gen_sounds.py
+"""
+import math
+import os
+import struct
+import wave
+
+SR = 44100
+OUT = os.path.join(os.path.dirname(__file__), "..", "web", "src", "assets", "sounds")
+
+
+def env(n, attack, decay, total):
+    """Smooth attack + exponential decay envelope over `total` seconds."""
+    a = max(1, int(attack * SR))
+    out = []
+    for i in range(n):
+        t = i / SR
+        if i < a:
+            amp = i / a
+        else:
+            amp = math.exp(-(t - attack) / decay)
+        out.append(amp)
+    return out
+
+
+def sine(freq_at, n):
+    """Sine with a per-sample frequency function freq_at(t)->Hz (phase-accurate)."""
+    out = []
+    phase = 0.0
+    for i in range(n):
+        t = i / SR
+        f = freq_at(t)
+        phase += 2 * math.pi * f / SR
+        out.append(math.sin(phase))
+    return out
+
+
+def mix(*layers):
+    n = max(len(l) for l in layers)
+    out = [0.0] * n
+    for l in layers:
+        for i, v in enumerate(l):
+            out[i] += v
+    return out
+
+
+def soft_clip(s):
+    return [math.tanh(v * 1.2) for v in s]
+
+
+def normalize(s, peak=0.9):
+    m = max(1e-9, max(abs(v) for v in s))
+    g = peak / m
+    return [v * g for v in s]
+
+
+def write_wav(name, samples):
+    samples = soft_clip(samples)
+    samples = normalize(samples, 0.85)
+    # 4ms fade-out tail so nothing clicks at the end.
+    tail = int(0.004 * SR)
+    for i in range(tail):
+        samples[-1 - i] *= i / tail
+    os.makedirs(OUT, exist_ok=True)
+    path = os.path.join(OUT, name)
+    with wave.open(path, "w") as w:
+        w.setnchannels(1)
+        w.setsampwidth(2)
+        w.setframerate(SR)
+        frames = b"".join(struct.pack("<h", int(max(-1, min(1, v)) * 32767)) for v in samples)
+        w.writeframes(frames)
+    print(f"wrote {name}  ({len(samples)/SR*1000:.0f} ms, {len(frames)} bytes)")
+
+
+def apply_env(sig, e):
+    return [s * a for s, a in zip(sig, e)]
+
+
+def bubble_pop():
+    """A cute bubble 'bloop' — pitch swoops up fast then a rounded body."""
+    dur = 0.16
+    n = int(dur * SR)
+    # Rising swoop is the signature of a bubble.
+    body = sine(lambda t: 360 + 720 * (1 - math.exp(-t * 38)), n)
+    body = apply_env(body, env(n, 0.004, 0.05, dur))
+    # A soft octave shimmer on top.
+    shimmer = sine(lambda t: 1080 + 400 * (1 - math.exp(-t * 38)), n)
+    shimmer = apply_env(shimmer, env(n, 0.003, 0.03, dur))
+    shimmer = [v * 0.25 for v in shimmer]
+    return mix(body, shimmer)
+
+
+def droplet():
+    """A clean little water-drop ping — high, downward, watery tail."""
+    dur = 0.22
+    n = int(dur * SR)
+    main = sine(lambda t: 1500 * math.exp(-t * 3.2) + 760, n)
+    main = apply_env(main, env(n, 0.003, 0.06, dur))
+    # Tiny resonant echo for a wet feel.
+    echo = sine(lambda t: 980, n)
+    echo = apply_env(echo, env(n, 0.05, 0.07, dur))
+    echo = [v * 0.2 for v in echo]
+    return mix(main, droop_delay(echo, 0.04))
+
+
+def droop_delay(sig, delay_s):
+    d = int(delay_s * SR)
+    return [0.0] * d + sig
+
+
+def bell(freq, dur, partials=((1, 1.0), (2.01, 0.5), (2.78, 0.28), (4.1, 0.12))):
+    """A soft inharmonic bell tone (sum of slightly detuned partials)."""
+    n = int(dur * SR)
+    layers = []
+    for ratio, amp in partials:
+        s = sine(lambda t, f=freq * ratio: f, n)
+        s = apply_env(s, env(n, 0.005, dur * 0.45, dur))
+        layers.append([v * amp for v in s])
+    return mix(*layers)
+
+
+def chime():
+    """Two gentle bell notes a soft fifth apart."""
+    a = bell(784, 0.42)  # G5
+    b = bell(1175, 0.40)  # D6
+    b = droop_delay(b, 0.10)
+    return mix(a, [v * 0.85 for v in b])
+
+
+def shimmer():
+    """Three quick ascending sparkles — glockenspiel-ish."""
+    notes = [988, 1319, 1976]  # B5, E6, B6
+    layers = []
+    for i, f in enumerate(notes):
+        s = bell(f, 0.30, partials=((1, 1.0), (2.7, 0.3), (5.1, 0.1)))
+        layers.append(droop_delay([v * 0.8 for v in s], 0.06 * i))
+    return mix(*layers)
+
+
+def cheer():
+    """A happy little major arpeggio C5-E5-G5-C6 on bells."""
+    notes = [523, 659, 784, 1047]
+    layers = []
+    for i, f in enumerate(notes):
+        s = bell(f, 0.36)
+        layers.append(droop_delay([v * 0.8 for v in s], 0.075 * i))
+    return mix(*layers)
+
+
+if __name__ == "__main__":
+    write_wav("pop.wav", bubble_pop())
+    write_wav("droplet.wav", droplet())
+    write_wav("chime.wav", chime())
+    write_wav("shimmer.wav", shimmer())
+    write_wav("cheer.wav", cheer())
+    print("done →", os.path.normpath(OUT))

web/package-lock.json

@@ -9,7 +9,15 @@
       "version": "0.0.0",
       "dependencies": {
         "@tiptap/extension-character-count": "^2.11.5",
+        "@tiptap/extension-color": "^2.27.2",
+        "@tiptap/extension-highlight": "^2.27.2",
+        "@tiptap/extension-image": "^2.27.2",
+        "@tiptap/extension-link": "^2.27.2",
         "@tiptap/extension-placeholder": "^2.11.5",
+        "@tiptap/extension-table": "^2.27.2",
+        "@tiptap/extension-table-cell": "^2.27.2",
+        "@tiptap/extension-table-header": "^2.27.2",
+        "@tiptap/extension-table-row": "^2.27.2",
         "@tiptap/extension-text-align": "^2.11.5",
         "@tiptap/extension-underline": "^2.11.5",
         "@tiptap/pm": "^2.11.5",
@@ -1568,6 +1576,20 @@
         "@tiptap/pm": "^2.7.0"
       }
     },
+    "node_modules/@tiptap/extension-color": {
+      "version": "2.27.2",
+      "resolved": "https://registry.npmjs.org/@tiptap/extension-color/-/extension-color-2.27.2.tgz",
+      "integrity": "sha512-sOKCP8/2V3sRM3FdWgMe1lFE5ewsWNCRafiVoujS1+TTHGCj4jw6W+LiumBUk7cRI8kXW/rqGWVC4RVdknYUCA==",
+      "license": "MIT",
+      "funding": {
+        "type": "github",
+        "url": "https://github.com/sponsors/ueberdosis"
+      },
+      "peerDependencies": {
+        "@tiptap/core": "^2.7.0",
+        "@tiptap/extension-text-style": "^2.7.0"
+      }
+    },
     "node_modules/@tiptap/extension-document": {
       "version": "2.27.2",
       "resolved": "https://registry.npmjs.org/@tiptap/extension-document/-/extension-document-2.27.2.tgz",
@@ -1652,6 +1674,19 @@
         "@tiptap/core": "^2.7.0"
       }
     },
+    "node_modules/@tiptap/extension-highlight": {
+      "version": "2.27.2",
+      "resolved": "https://registry.npmjs.org/@tiptap/extension-highlight/-/extension-highlight-2.27.2.tgz",
+      "integrity": "sha512-ZjlktDdMjruMJFAVz0TbQf0v92Jqkc7Ri1iZJqBXuLid+r+GxUzl2CVAV7qq5yagkGQgvAG+WGsMk880HgR3MA==",
+      "license": "MIT",
+      "funding": {
+        "type": "github",
+        "url": "https://github.com/sponsors/ueberdosis"
+      },
+      "peerDependencies": {
+        "@tiptap/core": "^2.7.0"
+      }
+    },
     "node_modules/@tiptap/extension-history": {
       "version": "2.27.2",
       "resolved": "https://registry.npmjs.org/@tiptap/extension-history/-/extension-history-2.27.2.tgz",
@@ -1680,6 +1715,19 @@
         "@tiptap/pm": "^2.7.0"
       }
     },
+    "node_modules/@tiptap/extension-image": {
+      "version": "2.27.2",
+      "resolved": "https://registry.npmjs.org/@tiptap/extension-image/-/extension-image-2.27.2.tgz",
+      "integrity": "sha512-5zL/BY41FIt72azVrCrv3n+2YJ/JyO8wxCcA4Dk1eXIobcgVyIdo4rG39gCqIOiqziAsqnqoj12QHTBtHsJ6mQ==",
+      "license": "MIT",
+      "funding": {
+        "type": "github",
+        "url": "https://github.com/sponsors/ueberdosis"
+      },
+      "peerDependencies": {
+        "@tiptap/core": "^2.7.0"
+      }
+    },
     "node_modules/@tiptap/extension-italic": {
       "version": "2.27.2",
       "resolved": "https://registry.npmjs.org/@tiptap/extension-italic/-/extension-italic-2.27.2.tgz",
@@ -1693,6 +1741,23 @@
         "@tiptap/core": "^2.7.0"
       }
     },
+    "node_modules/@tiptap/extension-link": {
+      "version": "2.27.2",
+      "resolved": "https://registry.npmjs.org/@tiptap/extension-link/-/extension-link-2.27.2.tgz",
+      "integrity": "sha512-bnP61qkr0Kj9Cgnop1hxn2zbOCBzNtmawxr92bVTOE31fJv6FhtCnQiD6tuPQVGMYhcmAj7eihtvuEMFfqEPcQ==",
+      "license": "MIT",
+      "dependencies": {
+        "linkifyjs": "^4.3.2"
+      },
+      "funding": {
+        "type": "github",
+        "url": "https://github.com/sponsors/ueberdosis"
+      },
+      "peerDependencies": {
+        "@tiptap/core": "^2.7.0",
+        "@tiptap/pm": "^2.7.0"
+      }
+    },
     "node_modules/@tiptap/extension-list-item": {
       "version": "2.27.2",
       "resolved": "https://registry.npmjs.org/@tiptap/extension-list-item/-/extension-list-item-2.27.2.tgz",
@@ -1759,6 +1824,59 @@
         "@tiptap/core": "^2.7.0"
       }
     },
+    "node_modules/@tiptap/extension-table": {
+      "version": "2.27.2",
+      "resolved": "https://registry.npmjs.org/@tiptap/extension-table/-/extension-table-2.27.2.tgz",
+      "integrity": "sha512-pDbhOpT5phZkcsyPjGBQlXv0+0hmdrvqHJ+dJjkGcCtlfy2pHiEIhmIItOFagc7wXy8G9iUFZ9Jie4zvDf+brg==",
+      "license": "MIT",
+      "funding": {
+        "type": "github",
+        "url": "https://github.com/sponsors/ueberdosis"
+      },
+      "peerDependencies": {
+        "@tiptap/core": "^2.7.0",
+        "@tiptap/pm": "^2.7.0"
+      }
+    },
+    "node_modules/@tiptap/extension-table-cell": {
+      "version": "2.27.2",
+      "resolved": "https://registry.npmjs.org/@tiptap/extension-table-cell/-/extension-table-cell-2.27.2.tgz",
+      "integrity": "sha512-9Lk46MjZMFzVZfOj9Kd7VgC6Odt6vmEhlCYVumErShUY7EkFqCw3b2IYoUtQkntfOEx/Afnhff/okNQwPsJeUA==",
+      "license": "MIT",
+      "funding": {
+        "type": "github",
+        "url": "https://github.com/sponsors/ueberdosis"
+      },
+      "peerDependencies": {
+        "@tiptap/core": "^2.7.0"
+      }
+    },
+    "node_modules/@tiptap/extension-table-header": {
+      "version": "2.27.2",
+      "resolved": "https://registry.npmjs.org/@tiptap/extension-table-header/-/extension-table-header-2.27.2.tgz",
+      "integrity": "sha512-ZEb6lbG0NbbodWLV0b4BS/QrDIPlUbCcuOsUxzqVvlMUY1Vg6Fj6fKwLaBcsIUDHi8sxZDBEgYEDw3BR/zcO6A==",
+      "license": "MIT",
+      "funding": {
+        "type": "github",
+        "url": "https://github.com/sponsors/ueberdosis"
+      },
+      "peerDependencies": {
+        "@tiptap/core": "^2.7.0"
+      }
+    },
+    "node_modules/@tiptap/extension-table-row": {
+      "version": "2.27.2",
+      "resolved": "https://registry.npmjs.org/@tiptap/extension-table-row/-/extension-table-row-2.27.2.tgz",
+      "integrity": "sha512-Nw9+tA56Y5HtLVP01NGCZSUuTQhJPtfK9OfmDgGgcxynn2cRVdEtj+9FNZqRhQ1iRVaAI+Rd4xRvX9qYePMOxw==",
+      "license": "MIT",
+      "funding": {
+        "type": "github",
+        "url": "https://github.com/sponsors/ueberdosis"
+      },
+      "peerDependencies": {
+        "@tiptap/core": "^2.7.0"
+      }
+    },
     "node_modules/@tiptap/extension-text": {
       "version": "2.27.2",
       "resolved": "https://registry.npmjs.org/@tiptap/extension-text/-/extension-text-2.27.2.tgz",
@@ -2828,6 +2946,12 @@
         "uc.micro": "^2.0.0"
       }
     },
+    "node_modules/linkifyjs": {
+      "version": "4.3.3",
+      "resolved": "https://registry.npmjs.org/linkifyjs/-/linkifyjs-4.3.3.tgz",
+      "integrity": "sha512-P8aEP5U/D1/IlTY2OeYsErdwh9bGuLE30NcXtKEjgdHcahveQoQwM2yZNsioQHsWFz0P7KKudisbrzCgR0sDHg==",
+      "license": "MIT"
+    },
     "node_modules/lottie-web": {
       "version": "5.13.0",
       "resolved": "https://registry.npmjs.org/lottie-web/-/lottie-web-5.13.0.tgz",

web/package.json

@@ -12,7 +12,15 @@
   },
   "dependencies": {
     "@tiptap/extension-character-count": "^2.11.5",
+    "@tiptap/extension-color": "^2.27.2",
+    "@tiptap/extension-highlight": "^2.27.2",
+    "@tiptap/extension-image": "^2.27.2",
+    "@tiptap/extension-link": "^2.27.2",
     "@tiptap/extension-placeholder": "^2.11.5",
+    "@tiptap/extension-table": "^2.27.2",
+    "@tiptap/extension-table-cell": "^2.27.2",
+    "@tiptap/extension-table-header": "^2.27.2",
+    "@tiptap/extension-table-row": "^2.27.2",
     "@tiptap/extension-text-align": "^2.11.5",
     "@tiptap/extension-underline": "^2.11.5",
     "@tiptap/pm": "^2.11.5",

web/src/App.tsx

@@ -13,6 +13,8 @@ import { StatusBar } from './components/StatusBar/StatusBar'
 import { PetalCompanion } from './components/Companion/PetalCompanion'
 import { UpdateBanner } from './components/UpdateBanner/UpdateBanner'
 import { useVersionWatch } from './hooks/useVersionWatch'
+import { PetalFall } from './effects/PetalFall'
+import { playSuggestionSound } from './audio/sounds'
 
 export default function App() {
   const updateAvailable = useVersionWatch()
@@ -190,6 +192,38 @@ export default function App() {
     setDocText('')
   }, [saveNow, isBlankDraft])
 
+  // Duplicate a document: copy its body/tone into a fresh doc titled "… (副本)",
+  // then open the copy. Tags aren't carried over (a fresh start for the copy).
+  const handleDuplicate = useCallback(
+    async (id: string) => {
+      try {
+        await saveNow() // flush in case we're duplicating the open doc
+        const src = await api.getDoc(id)
+        const fresh = await api.createDoc()
+        const dupTitle = `${src.title?.trim() || 'Untitled'} (副本)`
+        const updated = await api.updateDoc(fresh.id, {
+          title: dupTitle,
+          content: src.content,
+          content_text: src.content_text,
+          tone: src.tone,
+          word_count: src.word_count,
+        })
+        setDocs((prev) => [
+          { id: updated.id, title: updated.title, word_count: updated.word_count, updated_at: updated.updated_at, tags: [] },
+          ...prev,
+        ])
+        setCurrentDoc(updated)
+        setTitle(updated.title)
+        setWordCount(updated.word_count)
+        setTone(updated.tone || 'general')
+        setDocText(updated.content_text)
+      } catch (err) {
+        console.error('duplicate failed', err)
+      }
+    },
+    [saveNow],
+  )
+
   const handleDelete = useCallback(
     async (id: string) => {
       await api.deleteDoc(id)
@@ -309,8 +343,34 @@ export default function App() {
     [removeSuggestion],
   )
 
+  // Play a soft sound when freshly-checked suggestions arrive — one per distinct
+  // new type, lightly staggered so a batch reads as a little melody rather than
+  // a pile-up. We track which ids we've already chimed for, and only chime for
+  // recently-created suggestions so opening a doc with old pending advice stays
+  // silent (the existing set was created in a past session).
+  const chimedRef = useRef<Set<string>>(new Set())
+  useEffect(() => {
+    const fresh = suggestions.filter((s) => !chimedRef.current.has(s.id))
+    fresh.forEach((s) => chimedRef.current.add(s.id))
+    const justMade = fresh.filter(
+      (s) => Date.now() - new Date(s.created_at).getTime() < 12_000,
+    )
+    if (justMade.length === 0) return
+    const types = [...new Set(justMade.map((s) => s.type))].slice(0, 3)
+    const timers = types.map((t, i) =>
+      window.setTimeout(() => playSuggestionSound(t), i * 150),
+    )
+    return () => timers.forEach((id) => window.clearTimeout(id))
+  }, [suggestions])
+
+  // Forget chimed ids when switching documents so the set can't grow unbounded.
+  useEffect(() => {
+    chimedRef.current = new Set()
+  }, [currentDoc?.id])
+
   return (
     <div className="flex h-full flex-col">
+      <PetalFall />
       <header
         onMouseDown={handleChromeDown}
         className="petal-no-print flex h-12 shrink-0 items-center gap-2 px-5"
@@ -343,6 +403,7 @@ export default function App() {
             onSelect={openDoc}
             onCreate={handleCreate}
             onDelete={handleDelete}
+            onDuplicate={handleDuplicate}
             onToggleTag={handleToggleTag}
             onCreateTag={handleCreateTag}
           />
@@ -447,6 +508,7 @@ export default function App() {
         <PetalCompanion
           wordCount={wordCount}
           saveStatus={status}
+          llmDown={llmDown}
           editTick={editTick}
           acceptTick={acceptTick}
           text={docText}

web/src/api/client.ts

@@ -65,6 +65,7 @@ export interface WordMeaning {
 export interface WordInfo {
   word: string
   gloss: string // Chinese translation; '' when the word isn't in the gloss set
+  phonetic: string // IPA for the English word; '' when absent
   definitions: WordMeaning[]
   synonyms: string[]
 }
@@ -164,6 +165,10 @@ export const api = {
   exportUrl: (id: string, format: ExportFormat) =>
     `/api/docs/${id}/export?format=${format}`,
 
+  // Download URL for a whole-corpus backup: a zip of every document rendered in
+  // the given format. A one-click "download all my writing" safety net.
+  exportAllUrl: (format: ExportFormat) => `/api/docs/export-all?format=${format}`,
+
   // Offline word lookup (gloss + definition + synonyms) for the right-click popover.
   lookupWord: (word: string) => req<WordInfo>(`/word/${encodeURIComponent(word)}`),
   // Lightweight Chinese-only gloss for the inline hover/select tooltip — instant
@@ -182,6 +187,21 @@ export const api = {
   // highlighted snippet. Empty query returns []. Encodes the term for the URL.
   search: (q: string) => req<SearchResult[]>(`/search?q=${encodeURIComponent(q)}`),
 
+  // Image upload: posts a single image file as multipart form data and returns
+  // its served URL (e.g. /api/images/<hash>.png). Stored on disk by the binary,
+  // so the document JSON keeps a small URL reference instead of inlined base64.
+  // Doesn't go through req() because the body is FormData, not JSON.
+  uploadImage: async (file: File): Promise<{ url: string }> => {
+    const form = new FormData()
+    form.append('image', file)
+    const res = await fetch('/api/images', { method: 'POST', body: form })
+    if (!res.ok) {
+      const detail = await res.text().catch(() => '')
+      throw new Error(`${res.status} ${res.statusText}${detail ? `: ${detail}` : ''}`)
+    }
+    return res.json() as Promise<{ url: string }>
+  },
+
   // Tags. listTags returns the roster with per-tag document counts; createTag is
   // idempotent on name (returns the existing tag if it already exists);
   // updateTag recolors/renames; deleteTag removes it (assignments cascade).

web/src/audio/sounds.ts

@@ -0,0 +1,194 @@
+// Petal's cute sound palette. These are hand-picked mp3 effects (warm bubble
+// pops, a wooden "tok", rolling baoding balls, a Chinese fanfare for milestones,
+// and a playful "haiya" for errors) living as assets in ./assets/sounds. Vite
+// fingerprints and bundles them into dist, so they ride along inside the single
+// Go binary; here we just fetch + decode them once and play them through a Web
+// Audio bus with a soft master volume. The whole thing is opt-out-able and the
+// mute choice persists in localStorage so it survives reloads.
+
+import pop1Url from '../assets/sounds/pop1.mp3'
+import pop2Url from '../assets/sounds/pop2.mp3'
+import tokUrl from '../assets/sounds/tok.mp3'
+import blockUrl from '../assets/sounds/block.mp3'
+import baodingUrl from '../assets/sounds/baoding.mp3'
+import milestoneUrl from '../assets/sounds/milestone.mp3'
+import errorUrl from '../assets/sounds/error.mp3'
+
+const STORAGE_KEY = 'petal.sound'
+
+// Master volume — deliberately gentle. These are background delights, not alerts.
+const MASTER_GAIN = 0.5
+
+export type SoundName =
+  | 'pop1' // a soft bubble pop
+  | 'pop2' // a second bubble pop, slightly different
+  | 'tok' // a light wooden "tok"
+  | 'block' // a quick Chinese wood-block tap
+  | 'baoding' // rolling baoding-ball shimmer
+  | 'milestone' // a happy Chinese fanfare (word-count milestones)
+  | 'error' // a playful "haiya" — something went wrong
+
+const SOURCES: Record<SoundName, string> = {
+  pop1: pop1Url,
+  pop2: pop2Url,
+  tok: tokUrl,
+  block: blockUrl,
+  baoding: baodingUrl,
+  milestone: milestoneUrl,
+  error: errorUrl,
+}
+
+// The pool of light "something happened" notification sounds. We rotate through
+// these (skipping the one just played) so the same blip never repeats back to
+// back — a batch of suggestions or a run of tips stays lively instead of
+// hammering one note. milestone/error are deliberately NOT in the pool: they're
+// reserved for their specific moments.
+const POP_POOL: SoundName[] = ['pop1', 'pop2', 'tok', 'block', 'baoding']
+let lastPop = -1
+
+let ctx: AudioContext | null = null
+let master: GainNode | null = null
+const buffers = new Map<SoundName, AudioBuffer>()
+let enabled = readEnabled()
+const listeners = new Set<(on: boolean) => void>()
+
+function readEnabled(): boolean {
+  try {
+    return localStorage.getItem(STORAGE_KEY) !== 'off'
+  } catch {
+    return true
+  }
+}
+
+export function isSoundEnabled(): boolean {
+  return enabled
+}
+
+export function setSoundEnabled(on: boolean): void {
+  enabled = on
+  try {
+    localStorage.setItem(STORAGE_KEY, on ? 'on' : 'off')
+  } catch {
+    /* private mode — choice just won't persist */
+  }
+  listeners.forEach((fn) => fn(on))
+  // Touching the context on enable doubles as a user-gesture unlock + warm-up.
+  if (on) void ensureContext()
+}
+
+export function onSoundEnabledChange(fn: (on: boolean) => void): () => void {
+  listeners.add(fn)
+  return () => listeners.delete(fn)
+}
+
+// Lazily create the AudioContext (browsers require a user gesture before audio
+// can play; the first play() after a click will succeed, earlier ones resume).
+async function ensureContext(): Promise<AudioContext | null> {
+  if (typeof window === 'undefined') return null
+  const AC: typeof AudioContext | undefined =
+    window.AudioContext ??
+    (window as unknown as { webkitAudioContext?: typeof AudioContext }).webkitAudioContext
+  if (!AC) return null
+  if (!ctx) {
+    ctx = new AC()
+    master = ctx.createGain()
+    master.gain.value = MASTER_GAIN
+    // A gentle soft-clip so stacked sounds never crackle.
+    const shaper = ctx.createWaveShaper()
+    shaper.curve = softClipCurve()
+    master.connect(shaper)
+    shaper.connect(ctx.destination)
+  }
+  if (ctx.state === 'suspended') {
+    try {
+      await ctx.resume()
+    } catch {
+      /* will retry on the next gesture */
+    }
+  }
+  return ctx
+}
+
+function softClipCurve(): Float32Array<ArrayBuffer> {
+  const n = 1024
+  const curve = new Float32Array(new ArrayBuffer(n * 4))
+  for (let i = 0; i < n; i++) {
+    const x = (i / (n - 1)) * 2 - 1
+    curve[i] = Math.tanh(x * 1.2)
+  }
+  return curve
+}
+
+// Fetch + decode one sound, caching the decoded buffer for instant replay.
+async function load(name: SoundName, audio: AudioContext): Promise<AudioBuffer | null> {
+  const cached = buffers.get(name)
+  if (cached) return cached
+  try {
+    const res = await fetch(SOURCES[name])
+    const bytes = await res.arrayBuffer()
+    const buf = await audio.decodeAudioData(bytes)
+    buffers.set(name, buf)
+    return buf
+  } catch {
+    return null // asset missing / decode failed — sound just won't play
+  }
+}
+
+// Warm the cache so the first real play is instant.
+async function prime(): Promise<void> {
+  const audio = await ensureContext()
+  if (!audio) return
+  await Promise.all((Object.keys(SOURCES) as SoundName[]).map((n) => load(n, audio)))
+}
+
+// Fire a sound. No-op when muted or audio is unavailable. Never throws — a
+// delight that fails should fail silently.
+export function playSound(name: SoundName): void {
+  if (!enabled) return
+  void (async () => {
+    const audio = await ensureContext()
+    if (!audio || !master || audio.state !== 'running') return
+    const buf = await load(name, audio)
+    if (!buf) return
+    try {
+      const src = audio.createBufferSource()
+      src.buffer = buf
+      src.connect(master)
+      src.start()
+    } catch {
+      /* audio glitch — ignore */
+    }
+  })()
+}
+
+// Play the next notification pop from the pool, never repeating the last one, so
+// consecutive blips always sound different. This is the everyday "something
+// arrived" voice for suggestions, tips, and gentle cheers.
+export function playPop(): void {
+  let i = lastPop
+  // Tiny pool, so a short scan to pick anything but the last is plenty.
+  for (let tries = 0; tries < POP_POOL.length && i === lastPop; tries++) {
+    i = Math.floor(Math.random() * POP_POOL.length)
+  }
+  lastPop = i
+  playSound(POP_POOL[i])
+}
+
+// Each freshly-surfaced suggestion just gets a rotating pop — the variety comes
+// from the pool, not from the suggestion type, so a batch of same-type fixes
+// still sounds lively.
+export function playSuggestionSound(_type: string): void {
+  playPop()
+}
+
+// Unlock + warm audio on the first user gesture so the very first sound isn't
+// dropped by the browser's autoplay policy. Self-removing.
+if (typeof window !== 'undefined') {
+  const unlock = () => {
+    void prime()
+    window.removeEventListener('pointerdown', unlock)
+    window.removeEventListener('keydown', unlock)
+  }
+  window.addEventListener('pointerdown', unlock)
+  window.addEventListener('keydown', unlock)
+}

web/src/audio/speech.ts

@@ -0,0 +1,104 @@
+// Read-aloud for an ESL writer: hear how an English word or passage sounds.
+//
+// Primary path is server-side neural TTS (Piper, proxied at /api/tts) — natural,
+// consistent across devices, and far better than the browser's built-in voices
+// (which fall back to robotic espeak on Chromium). If the server route is absent
+// (TTS disabled) or unreachable, we fall back to the browser's Web Speech API so
+// the buttons still do something. No model or network is strictly required.
+
+// speechSupported reports whether read-aloud can do anything at all. Audio
+// playback is universal, so as long as we can construct an Audio element OR the
+// Web Speech API exists, the buttons should show. The server path is tried at
+// call time and degrades on its own.
+export function speechSupported(): boolean {
+  if (typeof window === 'undefined') return false
+  return typeof window.Audio === 'function' || 'speechSynthesis' in window
+}
+
+// webSpeechSupported gates the fallback path specifically.
+function webSpeechSupported(): boolean {
+  return typeof window !== 'undefined' && 'speechSynthesis' in window && 'SpeechSynthesisUtterance' in window
+}
+
+// current holds the in-flight server-audio element and its object URL so a rapid
+// re-tap can stop and replace it (mirroring speechSynthesis.cancel()).
+let current: { audio: HTMLAudioElement; url: string } | null = null
+
+// requestSeq increments on every speak() call so a slow fetch that resolves after
+// a newer tap can detect it's stale and bow out instead of double-playing.
+let requestSeq = 0
+
+function stopCurrent(): void {
+  if (current) {
+    current.audio.pause()
+    URL.revokeObjectURL(current.url)
+    current = null
+  }
+  if (webSpeechSupported()) window.speechSynthesis.cancel()
+}
+
+// pickVoice prefers an exact locale match (e.g. en-US), then any voice in the
+// same language family (en-*). Returns undefined to let the engine default.
+function pickVoice(lang: string): SpeechSynthesisVoice | undefined {
+  const voices = window.speechSynthesis.getVoices()
+  const base = lang.split('-')[0]
+  return voices.find((v) => v.lang === lang) || voices.find((v) => v.lang?.startsWith(base))
+}
+
+// speakWebSpeech is the fallback: the browser's built-in synthesizer. A touch
+// slower than default so learners can follow along.
+function speakWebSpeech(text: string, lang: string): void {
+  if (!webSpeechSupported()) return
+  const synth = window.speechSynthesis
+  synth.cancel()
+  const utterance = new SpeechSynthesisUtterance(text)
+  utterance.lang = lang
+  const voice = pickVoice(lang)
+  if (voice) utterance.voice = voice
+  utterance.rate = 0.95
+  synth.speak(utterance)
+}
+
+// speak reads `text` aloud, cancelling anything already in flight so rapid taps
+// don't queue up. `lang` defaults to US English (the language she's learning);
+// pass a zh locale to hear a Chinese passage. It tries the server's neural voice
+// first and silently falls back to the browser voice if that's unavailable (route
+// off, network error, or a 404 for a language with no configured voice).
+export function speak(text: string, lang = 'en-US'): void {
+  if (!text.trim()) return
+  stopCurrent()
+  const seq = ++requestSeq
+
+  fetch('/api/tts', {
+    method: 'POST',
+    headers: { 'Content-Type': 'application/json' },
+    body: JSON.stringify({ text, lang }),
+  })
+    .then((res) => {
+      if (!res.ok) throw new Error(`tts ${res.status}`)
+      return res.blob()
+    })
+    .then((blob) => {
+      // A newer tap superseded this one while the fetch was in flight — drop it.
+      if (seq !== requestSeq) return
+      const url = URL.createObjectURL(blob)
+      const audio = new Audio(url)
+      current = { audio, url }
+      // Release the blob once playback finishes (or errors) to avoid leaking.
+      const cleanup = () => {
+        if (current?.audio === audio) {
+          URL.revokeObjectURL(url)
+          current = null
+        }
+      }
+      audio.addEventListener('ended', cleanup)
+      audio.addEventListener('error', cleanup)
+      return audio.play()
+    })
+    .catch(() => {
+      // Server TTS unavailable for this request — use the browser voice instead,
+      // unless a newer tap has already superseded this one.
+      if (seq !== requestSeq) return
+      speakWebSpeech(text, lang)
+    })
+}

web/src/components/Companion/LottiePlayer.tsx

@@ -11,6 +11,8 @@ interface Props {
   animationData?: object
   loop?: boolean
   className?: string
+  // Mirror the animation left↔right (for assets drawn facing the wrong way).
+  flip?: boolean
   fallback: React.ReactNode
 }
 
@@ -48,7 +50,7 @@ function fitToContent(anim: AnimationItem, svg: SVGSVGElement) {
 // A thin wrapper over lottie-web (framework-agnostic, no React peer deps, fully
 // offline). Reloads the animation whenever the data changes — moods swap by
 // passing a different `animationData`.
-export function LottiePlayer({ animationData, loop = true, className, fallback }: Props) {
+export function LottiePlayer({ animationData, loop = true, className, flip, fallback }: Props) {
   const ref = useRef<HTMLDivElement>(null)
   const anim = useRef<AnimationItem | null>(null)
 
@@ -74,5 +76,12 @@ export function LottiePlayer({ animationData, loop = true, className, fallback }
   }, [animationData, loop])
 
   if (!animationData) return <>{fallback}</>
-  return <div ref={ref} className={className} aria-hidden />
+  return (
+    <div
+      ref={ref}
+      className={className}
+      style={flip ? { transform: 'scaleX(-1)' } : undefined}
+      aria-hidden
+    />
+  )
 }

web/src/components/Companion/PetalCompanion.tsx

@@ -7,6 +7,7 @@ import { COMPANIONS, DEFAULT_COMPANION } from './companions'
 interface Props {
   wordCount: number
   saveStatus: SaveStatus
+  llmDown: boolean
   editTick: number
   acceptTick: number
   text: string
@@ -27,10 +28,11 @@ const STORAGE_KEY = 'petal.companion'
 // useCompanion and shows a Mandarin-first speech bubble for cheers, tips, and
 // break reminders. Clicking the mascot opens a picker to switch companions
 // (the choice persists in localStorage).
-export function PetalCompanion({ wordCount, saveStatus, editTick, acceptTick, text }: Props) {
+export function PetalCompanion({ wordCount, saveStatus, llmDown, editTick, acceptTick, text }: Props) {
   const { mood, bubble, dismiss, holdBubble, releaseBubble } = useCompanion({
     wordCount,
     saveStatus,
+    llmDown,
     editTick,
     acceptTick,
     text,
@@ -145,7 +147,7 @@ export function PetalCompanion({ wordCount, saveStatus, editTick, acceptTick, te
           onClick={dismiss}
           onMouseEnter={holdBubble}
           onMouseLeave={releaseBubble}
-          className="petal-bubble pointer-events-auto max-w-[260px] cursor-pointer p-3 pr-3.5"
+          className="petal-bubble pointer-events-auto max-w-[420px] cursor-pointer p-4 pr-5"
           style={{
             background: 'var(--color-surface)',
             border: '1px solid var(--color-border)',
@@ -156,10 +158,16 @@ export function PetalCompanion({ wordCount, saveStatus, editTick, acceptTick, te
           }}
           title="Click to dismiss"
         >
-          <p className="text-sm font-bold leading-snug" style={{ color: 'var(--color-plum)' }}>
+          <p
+            className="font-bold leading-snug"
+            style={{ color: 'var(--color-plum)', fontSize: '1.4rem' }}
+          >
             {bubble.zh}
           </p>
-          <p className="mt-0.5 text-xs leading-snug" style={{ color: 'var(--color-muted)' }}>
+          <p
+            className="mt-0.5 leading-snug"
+            style={{ color: 'var(--color-muted)', fontSize: '1.15rem' }}
+          >
             {bubble.en}
           </p>
         </div>
@@ -172,8 +180,9 @@ export function PetalCompanion({ wordCount, saveStatus, editTick, acceptTick, te
         aria-label="Choose a companion"
         className={`petal-companion pointer-events-auto select-none ${napping ? 'petal-companion-sleep' : ''}`}
         style={{
-          width: 144,
+          // Size scales with the viewport — see --petal-companion-size in index.css.
-          height: 144,
+          width: 'var(--petal-companion-size)',
+          height: 'var(--petal-companion-size)',
           padding: 0,
           borderRadius: 'var(--radius-pill)',
           background: 'var(--color-surface-alt)',
@@ -187,8 +196,13 @@ export function PetalCompanion({ wordCount, saveStatus, editTick, acceptTick, te
         <LottiePlayer
           key={companion.id}
           animationData={animationData}
-          className="h-32 w-32"
+          flip={companion.flip}
-          fallback={<span style={{ fontSize: 72, lineHeight: 1 }}>{MOOD_EMOJI[renderMood]}</span>}
+          className="petal-companion-art"
+          fallback={
+            <span style={{ fontSize: 'calc(var(--petal-companion-size) * 0.5)', lineHeight: 1 }}>
+              {MOOD_EMOJI[renderMood]}
+            </span>
+          }
         />
         {napping && (
           <span className="petal-zzz absolute" aria-hidden style={{ color: 'var(--color-muted)' }}>

web/src/components/Companion/companions.ts

@@ -1,6 +1,9 @@
 import type { Mood } from './useCompanion'
 import sleepingCat from './animations/sleeping-cat.json'
 import happyDog from './animations/happy-dog.json'
+import wiggleDog from './animations/wiggle-dog.json'
+import butterfly from './animations/butterfly.json'
+import parrot from './animations/parrot.json'
 
 export interface Companion {
   id: string
@@ -12,6 +15,9 @@ export interface Companion {
   // When true the mascot keeps its calm sway + drifting zzz regardless of mood
   // (e.g. a cat that's always asleep but still talks in its sleep).
   alwaysAsleep?: boolean
+  // When true the rendered animation is mirrored left↔right — for assets drawn
+  // facing the "wrong" way for our bottom-right corner (e.g. the parrot).
+  flip?: boolean
 }
 
 // The roster. Add a companion by dropping a pure-vector Lottie JSON into
@@ -44,6 +50,43 @@ export const COMPANIONS: Companion[] = [
       // no sleeping clip → stays in its idle pose instead of visibly napping
     },
   },
+  {
+    id: 'wiggle-dog',
+    name: 'Wiggle Dog',
+    zh: '摇尾狗',
+    emoji: '🐕',
+    animations: {
+      idle: wiggleDog,
+      happy: wiggleDog,
+      talking: wiggleDog,
+      celebrate: wiggleDog,
+    },
+  },
+  {
+    id: 'butterfly',
+    name: 'Butterfly',
+    zh: '蝴蝶',
+    emoji: '🦋',
+    animations: {
+      idle: butterfly,
+      happy: butterfly,
+      talking: butterfly,
+      celebrate: butterfly,
+    },
+  },
+  {
+    id: 'parrot',
+    name: 'Parrot',
+    zh: '鹦鹉',
+    emoji: '🦜',
+    flip: true, // asset faces left; mirror it to face into the page
+    animations: {
+      idle: parrot,
+      happy: parrot,
+      talking: parrot,
+      celebrate: parrot,
+    },
+  },
 ]
 
 export const DEFAULT_COMPANION = 'cat'

web/src/components/Companion/tips.ts

@@ -48,8 +48,16 @@ export const WELCOME_BACK: Line = {
   en: 'Welcome back ✨ let’s keep going!',
 }
 
-// Word-count milestones worth a little cheer.
+// Gentle "haiya, something went wrong" lines — paired with the error sound when
-export const MILESTONES = [50, 100, 250, 500, 1000, 2000]
+// the LLM is unreachable or a save fails. Never alarming, always reassuring.
+export const ERRORS: Line[] = [
+  { zh: '哎呀～出了点小问题，你的字都还在哦。', en: 'Oops — a little hiccup, but your words are safe.' },
+  { zh: '哎呀，我这边卡了一下，马上就好。', en: 'Haiya, I got stuck for a sec — back in a moment.' },
+  { zh: '别担心，等一下再试试看 🍵', en: "Don't worry — let's try again in a bit. 🍵" },
+]
+
+// Word-count milestones worth a little cheer — every 100 words, on up.
+export const MILESTONES = Array.from({ length: 100 }, (_, i) => (i + 1) * 100)
 
 export function milestoneLine(n: number): Line {
   return {

web/src/components/Companion/useCompanion.ts

@@ -3,6 +3,7 @@ import type { SaveStatus } from '../../hooks/useAutoSave'
 import {
   BREAKS,
   ENCOURAGEMENTS,
+  ERRORS,
   GREETING,
   MILESTONES,
   TIPS,
@@ -12,12 +13,13 @@ import {
   type Line,
 } from './tips'
 import { analyzeProse } from './prose'
+import { playPop, playSound, type SoundName } from '../../audio/sounds'
 
 // The kitten's expression. Maps to a Lottie animation when assets are present,
 // otherwise to an emoji placeholder (see PetalCompanion).
 export type Mood = 'idle' | 'happy' | 'talking' | 'sleeping' | 'celebrate'
 
-export type BubbleTone = 'cheer' | 'tip' | 'break'
+export type BubbleTone = 'cheer' | 'tip' | 'break' | 'error'
 export interface Bubble extends Line {
   tone: BubbleTone
 }
@@ -25,6 +27,9 @@ export interface Bubble extends Line {
 interface Signals {
   wordCount: number
   saveStatus: SaveStatus
+  // True while the writing-assist LLM is unreachable (timeout / down). A
+  // false→true flip earns a gentle "haiya".
+  llmDown: boolean
   // Monotonic counters bumped by the app on each editor change / accepted
   // suggestion — lets the companion react without a full event bus.
   editTick: number
@@ -42,10 +47,10 @@ const PROACTIVE_GAP = 40_000 // floor between any two unsolicited bubbles
 // How long a bubble lingers. These are *floors* — readBubbleMs extends them by
 // how much there is to read, since she reads both the Mandarin and the English
 // (and tips now quote a slice of her own sentence, so they run longer).
-const BUBBLE_MS = 9_000 // baseline for a tip / break
+const BUBBLE_MS = 14_000 // baseline for a tip / break
-const CHEER_MS = 6_000 // a short cheer still needs a beat in two languages
+const CHEER_MS = 9_000 // a short cheer still needs a beat in two languages
-const READ_MS_PER_CHAR = 45 // ~22 chars/sec, generous for a bilingual read
+const READ_MS_PER_CHAR = 55 // ~18 chars/sec, generous for a bilingual ESL read
-const MAX_BUBBLE_MS = 20_000 // cap so a long quote can't pin the bubble forever
+const MAX_BUBBLE_MS = 32_000 // cap so a long quote can't pin the bubble forever
 const HOVER_GRACE_MS = 3_000 // lingers this long after she stops hovering
 const now = () => Date.now()
 
@@ -61,7 +66,7 @@ function readBubbleMs(b: Bubble): number {
 // useCompanion is the behavior engine: it watches writing signals and decides
 // when the kitten speaks, what mood it shows, and how to pace itself so the
 // companion feels alive without interrupting. UI-agnostic — returns state only.
-export function useCompanion({ wordCount, saveStatus, editTick, acceptTick, text }: Signals) {
+export function useCompanion({ wordCount, saveStatus, llmDown, editTick, acceptTick, text }: Signals) {
   const [mood, setMood] = useState<Mood>('idle')
   const [bubble, setBubble] = useState<Bubble | null>(null)
 
@@ -87,7 +92,8 @@ export function useCompanion({ wordCount, saveStatus, editTick, acceptTick, text
 
   // Show a bubble + talking mood, then settle back. `proactive` messages respect
   // the spacing floor; user-triggered ones (cheers) always go through.
-  const say = useCallback((b: Bubble, opts?: { proactive?: boolean; celebrate?: boolean }) => {
+  const say = useCallback(
+    (b: Bubble, opts?: { proactive?: boolean; celebrate?: boolean; sound?: SoundName }) => {
     const t = now()
     if (opts?.proactive) {
       if (bubble) return
@@ -95,6 +101,11 @@ export function useCompanion({ wordCount, saveStatus, editTick, acceptTick, text
       lastProactive.current = t
     }
     sleeping.current = false
+    // A sound to match the bubble: callers can name a specific one (the milestone
+    // fanfare, the "haiya" on errors); everything else gets a rotating pop so the
+    // same blip never repeats back to back.
+    if (opts?.sound) playSound(opts.sound)
+    else playPop()
     setBubble(b)
     setMood(opts?.celebrate ? 'celebrate' : 'talking')
     clearTimeout(bubbleTimer.current)
@@ -190,11 +201,39 @@ export function useCompanion({ wordCount, saveStatus, editTick, acceptTick, text
       const n = MILESTONES[nextMilestone.current]
       nextMilestone.current += 1
       // Skip silently if we're just loading a long doc (no edits yet).
-      if (!firstEdit.current) say({ ...milestoneLine(n), tone: 'cheer' }, { celebrate: true })
+      if (!firstEdit.current)
+        say({ ...milestoneLine(n), tone: 'cheer' }, { celebrate: true, sound: 'milestone' })
     }
     // eslint-disable-next-line react-hooks/exhaustive-deps
   }, [wordCount])
 
+  // Trouble — a save failed or the writing-assist LLM went unreachable. Play the
+  // playful "haiya" and reassure her, but never spam: at most one per gap, and
+  // only on the *transition* into trouble (not while it lingers).
+  const lastError = useRef(0)
+  const ERROR_GAP = 20_000
+  const wasLlmDown = useRef(false)
+  const haiya = useCallback(() => {
+    const t = now()
+    if (t - lastError.current < ERROR_GAP) {
+      playSound('error') // still acknowledge it, just without a fresh bubble
+      return
+    }
+    lastError.current = t
+    say({ ...pick(ERRORS), tone: 'error' }, { sound: 'error' })
+  }, [say])
+
+  useEffect(() => {
+    if (saveStatus === 'error') haiya()
+    // eslint-disable-next-line react-hooks/exhaustive-deps
+  }, [saveStatus])
+
+  useEffect(() => {
+    if (llmDown && !wasLlmDown.current) haiya()
+    wasLlmDown.current = llmDown
+    // eslint-disable-next-line react-hooks/exhaustive-deps
+  }, [llmDown])
+
   // Occasional gentle cheer on a successful save (kept rare so it isn't noise).
   useEffect(() => {
     if (saveStatus === 'saved' && Math.random() < 0.18) {

web/src/components/DocList/DocList.tsx

@@ -1,5 +1,5 @@
 import { useMemo, useState } from 'react'
-import type { DocSummary, Tag, TagColor } from '../../api/client'
+import { api, type DocSummary, type Tag, type TagColor } from '../../api/client'
 import { DocListItem } from './DocListItem'
 import { SearchBox } from './SearchBox'
 import { TagChip } from './TagChip'
@@ -11,10 +11,19 @@ interface Props {
   onSelect: (id: string) => void
   onCreate: () => void
   onDelete: (id: string) => void
+  onDuplicate: (id: string) => void
   onToggleTag: (docId: string, tag: Tag) => void
   onCreateTag: (docId: string, name: string, color: TagColor) => void
 }
 
+// Sidebar sort orders. 'recent' keeps the server's updated_at-desc ordering.
+type SortMode = 'recent' | 'title' | 'longest'
+const SORTS: { value: SortMode; label: string }[] = [
+  { value: 'recent', label: '最近 · Recent' },
+  { value: 'title', label: '标题 · Title' },
+  { value: 'longest', label: '字数 · Longest' },
+]
+
 // DocList is the sidebar: a cross-document search box, a tag filter bar, the New
 // button, and the document browser (each row showing its tag chips).
 export function DocList({
@@ -24,6 +33,7 @@ export function DocList({
   onSelect,
   onCreate,
   onDelete,
+  onDuplicate,
   onToggleTag,
   onCreateTag,
 }: Props) {
@@ -31,11 +41,21 @@ export function DocList({
   // disappears from the roster.
   const [filterId, setFilterId] = useState<string | null>(null)
   const activeFilter = filterId && roster.some((t) => t.id === filterId) ? filterId : null
+  const [sort, setSort] = useState<SortMode>('recent')
 
-  const filtered = useMemo(
+  const filtered = useMemo(() => {
-    () => (activeFilter ? docs.filter((d) => d.tags.some((t) => t.id === activeFilter)) : docs),
+    const base = activeFilter ? docs.filter((d) => d.tags.some((t) => t.id === activeFilter)) : docs
-    [docs, activeFilter],
+    if (sort === 'recent') return base // already updated_at-desc from the server
-  )
+    const arr = [...base]
+    if (sort === 'title') {
+      arr.sort((a, b) =>
+        (a.title || 'Untitled').localeCompare(b.title || 'Untitled', undefined, { sensitivity: 'base' }),
+      )
+    } else {
+      arr.sort((a, b) => b.word_count - a.word_count)
+    }
+    return arr
+  }, [docs, activeFilter, sort])
 
   // Only surface tags that are actually in use, so the filter bar stays tidy.
   const usedTags = useMemo(() => roster.filter((t) => (t.doc_count ?? 0) > 0), [roster])
@@ -72,6 +92,25 @@ export function DocList({
         <span className="text-base leading-none">＋</span> New Doc
       </button>
 
+      {docs.length > 1 && (
+        <div className="flex items-center justify-end gap-1.5 px-1 text-xs" style={{ color: 'var(--color-muted)' }}>
+          <span aria-hidden>↕</span>
+          <select
+            aria-label="Sort documents"
+            value={sort}
+            onChange={(e) => setSort(e.target.value as SortMode)}
+            className="bg-transparent text-xs font-semibold focus:outline-none"
+            style={{ color: 'var(--color-plum)' }}
+          >
+            {SORTS.map((s) => (
+              <option key={s.value} value={s.value}>
+                {s.label}
+              </option>
+            ))}
+          </select>
+        </div>
+      )}
+
       <div className="flex flex-1 flex-col gap-0.5 overflow-y-auto">
         {filtered.length === 0 ? (
           <p className="px-3 py-6 text-center text-xs" style={{ color: 'var(--color-muted)' }}>
@@ -86,12 +125,30 @@ export function DocList({
               roster={roster}
               onSelect={() => onSelect(doc.id)}
               onDelete={() => onDelete(doc.id)}
+              onDuplicate={() => onDuplicate(doc.id)}
               onToggleTag={onToggleTag}
               onCreateTag={onCreateTag}
             />
           ))
         )}
       </div>
+
+      {/* Whole-corpus backup — "download all my writing". Plain download links so
+          the browser saves the zip; no extra app state needed. */}
+      {docs.length > 0 && (
+        <div
+          className="flex items-center gap-2 px-1 pt-1 text-xs"
+          style={{ color: 'var(--color-muted)', borderTop: '1px solid var(--color-border)' }}
+        >
+          <span className="font-semibold">备份 · Back up all:</span>
+          <a href={api.exportAllUrl('docx')} download className="font-bold hover:underline" style={{ color: 'var(--color-accent-hover)' }}>
+            Word
+          </a>
+          <a href={api.exportAllUrl('md')} download className="font-bold hover:underline" style={{ color: 'var(--color-accent-hover)' }}>
+            Markdown
+          </a>
+        </div>
+      )}
     </aside>
   )
 }

web/src/components/DocList/DocListItem.tsx

@@ -9,6 +9,7 @@ interface Props {
   roster: Tag[]
   onSelect: () => void
   onDelete: () => void
+  onDuplicate: () => void
   onToggleTag: (docId: string, tag: Tag) => void
   onCreateTag: (docId: string, name: string, color: TagColor) => void
 }
@@ -21,6 +22,7 @@ export function DocListItem({
   roster,
   onSelect,
   onDelete,
+  onDuplicate,
   onToggleTag,
   onCreateTag,
 }: Props) {
@@ -65,6 +67,19 @@ export function DocListItem({
         >
           🏷️
         </button>
+        <button
+          type="button"
+          aria-label="Duplicate document"
+          title="副本 · Duplicate"
+          onClick={(e) => {
+            e.stopPropagation()
+            onDuplicate()
+          }}
+          className="petal-row-action flex h-7 w-7 shrink-0 items-center justify-center text-sm"
+          style={{ borderRadius: 'var(--radius-pill)', color: 'var(--color-muted)' }}
+        >
+          ⧉
+        </button>
         <button
           type="button"
           aria-label="Delete document"

web/src/components/Editor/EditorCore.tsx

@@ -4,6 +4,17 @@ import Underline from '@tiptap/extension-underline'
 import TextAlign from '@tiptap/extension-text-align'
 import Placeholder from '@tiptap/extension-placeholder'
 import CharacterCount from '@tiptap/extension-character-count'
+import Link from '@tiptap/extension-link'
+import { Color } from '@tiptap/extension-color'
+import TextStyle from '@tiptap/extension-text-style'
+import Highlight from '@tiptap/extension-highlight'
+import Image from '@tiptap/extension-image'
+import Table from '@tiptap/extension-table'
+import TableRow from '@tiptap/extension-table-row'
+import TableHeader from '@tiptap/extension-table-header'
+import TableCell from '@tiptap/extension-table-cell'
+import { FontSize } from './FontSize'
+import type { EditorView } from '@tiptap/pm/view'
 import { useCallback, useEffect, useRef, useState } from 'react'
 import { Toolbar } from '../Toolbar/Toolbar'
 import { SuggestionCard } from './SuggestionCard'
@@ -13,8 +24,12 @@ import { MisspellCard } from './MisspellCard'
 import { WordCard } from './WordCard'
 import { GlossTip } from './GlossTip'
 import { SelectionBubble } from './SelectionBubble'
+import { SearchHighlight } from './SearchHighlight'
+import { FindReplace } from './FindReplace'
+import { Typography } from './Typography'
 import { RewritePreview, type RewriteStatus } from './RewritePreview'
 import { api, type Suggestion, type WordInfo } from '../../api/client'
+import { speak, speechSupported } from '../../audio/speech'
 import type { SpellChecker } from '../../hooks/useSpellChecker'
 
 export interface EditorChange {
@@ -103,6 +118,21 @@ function parseDoc(raw: string): object | undefined {
   return undefined
 }
 
+// uploadImageInto sends an image file to the store and inserts the returned URL
+// as an image node — at `pos` if given (a drop point), otherwise at the current
+// selection (a paste). Shared by the paste/drop handlers and the toolbar button.
+export async function uploadImageInto(view: EditorView, file: File, pos?: number) {
+  try {
+    const { url } = await api.uploadImage(file)
+    const { schema } = view.state
+    const node = schema.nodes.image.create({ src: url })
+    const at = pos ?? view.state.selection.from
+    view.dispatch(view.state.tr.insert(at, node))
+  } catch (err) {
+    console.error('image upload failed', err)
+  }
+}
+
 interface HoverState {
   suggestion: Suggestion
   top: number
@@ -181,22 +211,61 @@ export function EditorCore({
   // The rewrite affordances: a bubble over the current selection, and the
   // preview that replaces it once a style is chosen.
   const [selection, setSelection] = useState<SelectionState | null>(null)
+  // True while a pointer is held down (a drag-select in progress). The rewrite
+  // bubble stays hidden until release so it never appears — or re-anchors — under
+  // a still-moving cursor, which is where the user is trying to click.
+  const [dragging, setDragging] = useState(false)
   const [rewrite, setRewrite] = useState<RewriteState | null>(null)
   const rewriteReqRef = useRef(0)
+  // The in-document Find & Replace bar (Ctrl/Cmd+F).
+  const [findOpen, setFindOpen] = useState(false)
 
   const editor = useEditor({
     extensions: [
       StarterKit,
       Underline,
+      TextStyle,
+      Color,
+      FontSize,
+      Highlight.configure({ multicolor: true }),
+      Link.configure({ openOnClick: false, autolink: true, HTMLAttributes: { rel: 'noopener noreferrer nofollow' } }),
+      Image.configure({ inline: false, HTMLAttributes: { class: 'petal-image' } }),
+      Table.configure({ resizable: true, HTMLAttributes: { class: 'petal-table' } }),
+      TableRow,
+      TableHeader,
+      TableCell,
       TextAlign.configure({ types: ['heading', 'paragraph'] }),
       Placeholder.configure({ placeholder: 'Start writing…' }),
       CharacterCount,
       SuggestionHighlight,
       SpellCheck,
+      SearchHighlight,
+      Typography,
     ],
     content: parseDoc(initialContent),
     editorProps: {
       attributes: { class: 'petal-prose focus:outline-none' },
+      // Dropping or pasting an image file uploads it and inserts it at the drop
+      // point (or the cursor for a paste). Returns true to consume the event so
+      // ProseMirror doesn't also try to handle the raw file. Non-image pastes
+      // fall through to the default handler.
+      handleDrop: (view, event) => {
+        const files = (event as DragEvent).dataTransfer?.files
+        const image = files && Array.from(files).find((f) => f.type.startsWith('image/'))
+        if (!image) return false
+        event.preventDefault()
+        const coords = view.posAtCoords({ left: (event as DragEvent).clientX, top: (event as DragEvent).clientY })
+        uploadImageInto(view, image, coords?.pos)
+        return true
+      },
+      handlePaste: (view, event) => {
+        const files = event.clipboardData?.files
+        const image = files && Array.from(files).find((f) => f.type.startsWith('image/'))
+        if (!image) return false
+        event.preventDefault()
+        uploadImageInto(view, image)
+        return true
+      },
     },
     onFocus: () => onFocusMode?.(),
     onUpdate: ({ editor }) => {
@@ -410,20 +479,17 @@ export function EditorCore({
     setMisspell(null)
   }, [misspell, onAddWord])
 
-  // Right-click a word to look it up: resolve the exact word span under the
+  // openWordLookup resolves the exact word span at a document position, anchors a
-  // pointer, anchor a popover beneath it, and kick off the offline lookup. The
+  // popover beneath it, and kicks off the offline lookup. The card opens
-  // card opens immediately in a loading state and fills in when the (local)
+  // immediately in a loading state and fills in when the (local) lookup returns.
-  // lookup returns. Right-clicking off any word falls through to the native menu.
+  // Shared by right-click, the keyboard shortcut (caret), and touch long-press.
-  const handleContextMenu = useCallback(
+  const openWordLookup = useCallback(
-    (e: React.MouseEvent) => {
+    (pos: number) => {
       if (!editor) return
-      const coords = editor.view.posAtCoords({ left: e.clientX, top: e.clientY })
+      const range = wordAt(editor.state.doc, pos)
-      if (!coords) return
-      const range = wordAt(editor.state.doc, coords.pos)
       if (!range) return
       const wrapper = wrapperRef.current
       if (!wrapper) return
-      e.preventDefault()
       // Anchor under the word itself (not the click point) so the card lines up
       // with the text the way the spelling popover does.
       const start = editor.view.coordsAtPos(range.from)
@@ -454,6 +520,37 @@ export function EditorCore({
     [editor, closeCard],
   )
 
+  // Right-click a word to look it up. Right-clicking off any word falls through
+  // to the native menu (so copy/paste-by-menu still works — see the Selection fix).
+  const handleContextMenu = useCallback(
+    (e: React.MouseEvent) => {
+      if (!editor) return
+      const coords = editor.view.posAtCoords({ left: e.clientX, top: e.clientY })
+      if (!coords) return
+      if (!wordAt(editor.state.doc, coords.pos)) return
+      e.preventDefault()
+      openWordLookup(coords.pos)
+    },
+    [editor, openWordLookup],
+  )
+
+  // Touch has no hover or right-click, so a long-press (~500ms without moving)
+  // opens the word lookup at the finger — the touch equivalent of right-click.
+  const longPressTimer = useRef<ReturnType<typeof setTimeout>>(undefined)
+  const handleTouchStart = useCallback(
+    (e: React.TouchEvent) => {
+      if (!editor || e.touches.length !== 1) return
+      const { clientX, clientY } = e.touches[0]
+      clearTimeout(longPressTimer.current)
+      longPressTimer.current = setTimeout(() => {
+        const coords = editor.view.posAtCoords({ left: clientX, top: clientY })
+        if (coords) openWordLookup(coords.pos)
+      }, 500)
+    },
+    [editor, openWordLookup],
+  )
+  const cancelLongPress = useCallback(() => clearTimeout(longPressTimer.current), [])
+
   const replaceWord = useCallback(
     (synonym: string) => {
       if (editor && wordInfo) {
@@ -626,10 +723,52 @@ export function EditorCore({
     return () => document.removeEventListener('mousedown', onDown)
   }, [misspell])
 
+  // A pointer-down in the editor starts a (possible) drag-select, which keeps the
+  // rewrite bubble hidden until release. A pointer-down on the bubble itself is
+  // excluded so its buttons stay clickable. Release always re-arms the bubble.
+  const handlePointerDown = useCallback((e: React.PointerEvent) => {
+    if ((e.target as HTMLElement).closest('.petal-selection-bubble')) return
+    setDragging(true)
+  }, [])
+
+  useEffect(() => {
+    const onUp = () => setDragging(false)
+    document.addEventListener('pointerup', onUp)
+    return () => document.removeEventListener('pointerup', onUp)
+  }, [])
+
+  // Editor keyboard shortcuts, so the ESL helpers aren't mouse-only:
+  //   Ctrl/Cmd+F — Find & Replace (overrides the browser find, which can't see
+  //                into the editor's content anyway)
+  //   Ctrl/Cmd+D — look up the word at the caret (the keyboard "right-click")
+  //   Ctrl/Cmd+J — rewrite the current selection more naturally (✨更自然)
+  useEffect(() => {
+    if (!editor) return
+    const onKey = (e: KeyboardEvent) => {
+      if (!(e.ctrlKey || e.metaKey) || e.altKey) return
+      const k = e.key.toLowerCase()
+      if (k === 'f') {
+        e.preventDefault()
+        setFindOpen(true)
+      } else if (k === 'd') {
+        e.preventDefault()
+        openWordLookup(editor.state.selection.from)
+      } else if (k === 'j') {
+        if (!editor.state.selection.empty) {
+          e.preventDefault()
+          handleRewrite('natural')
+        }
+      }
+    }
+    window.addEventListener('keydown', onKey)
+    return () => window.removeEventListener('keydown', onKey)
+  }, [editor, openWordLookup, handleRewrite])
+
   useEffect(() => () => {
     clearTimeout(closeTimer.current)
     clearTimeout(confettiTimer.current)
     clearTimeout(glossTimer.current)
+    clearTimeout(longPressTimer.current)
   }, [])
 
   // While pinned (Ask Petal open), a pointer-down outside the card closes it —
@@ -667,16 +806,22 @@ export function EditorCore({
         onMouseOut={handleMouseOut}
         onMouseMove={handleMouseMove}
         onMouseLeave={handleMouseLeave}
+        onPointerDown={handlePointerDown}
         onClick={handleSpellClick}
         onContextMenu={handleContextMenu}
+        onTouchStart={handleTouchStart}
+        onTouchMove={cancelLongPress}
+        onTouchEnd={cancelLongPress}
       >
         <EditorContent editor={editor} className="h-full" />
+        {findOpen && editor && <FindReplace editor={editor} onClose={() => setFindOpen(false)} />}
         {confetti && <Confetti top={confetti.top} left={confetti.left} />}
         {gloss && <GlossTip gloss={gloss.gloss} style={{ top: gloss.top, left: gloss.left }} />}
-        {selection && !rewrite && (
+        {selection && !rewrite && !dragging && (
           <SelectionBubble
             style={{ top: selection.top, left: selection.left, transform: 'translateY(calc(-100% - 8px))' }}
             onRewrite={handleRewrite}
+            onSpeak={speechSupported() ? () => speak(selection.text) : null}
           />
         )}
         {rewrite && (

web/src/components/Editor/FindReplace.tsx

@@ -0,0 +1,238 @@
+import { useCallback, useEffect, useRef, useState } from 'react'
+import type { Editor } from '@tiptap/react'
+import { clearSearch, getSearchState, setActive, setSearch } from './SearchHighlight'
+
+// FindReplace is the in-document search bar (Ctrl/Cmd+F). It drives the
+// SearchHighlight decoration layer: typing updates the highlighted matches, the
+// arrows step through them (scrolling each into view), and replace / replace-all
+// edit the document in place. Bilingual, zh-first, matching the editor chrome.
+
+interface Props {
+  editor: Editor
+  onClose: () => void
+}
+
+const CJK = "'Nunito','PingFang SC','Microsoft YaHei','Noto Sans CJK SC',sans-serif"
+
+// Scroll the active match into the middle of the viewport without moving the
+// selection (which would otherwise pop the rewrite bubble).
+function scrollToActive(editor: Editor) {
+  const st = getSearchState(editor.state)
+  if (st.active < 0) return
+  const m = st.matches[st.active]
+  const dom = editor.view.domAtPos(m.from)
+  const el = dom.node.nodeType === Node.TEXT_NODE ? dom.node.parentElement : (dom.node as HTMLElement)
+  el?.scrollIntoView({ block: 'center', behavior: 'smooth' })
+}
+
+export function FindReplace({ editor, onClose }: Props) {
+  const [query, setQuery] = useState('')
+  const [replacement, setReplacement] = useState('')
+  const [caseSensitive, setCaseSensitive] = useState(false)
+  const [showReplace, setShowReplace] = useState(false)
+  // Mirror of the plugin's match count + active index for the "n / m" readout.
+  const [{ count, active }, setCounts] = useState({ count: 0, active: -1 })
+  const findRef = useRef<HTMLInputElement>(null)
+
+  const sync = useCallback(() => {
+    const st = getSearchState(editor.state)
+    setCounts({ count: st.matches.length, active: st.active })
+  }, [editor])
+
+  // Push the query into the decoration layer whenever it (or case-sensitivity)
+  // changes, then jump to the first match.
+  useEffect(() => {
+    setSearch(editor.state, editor.view.dispatch, query, caseSensitive)
+    sync()
+    scrollToActive(editor)
+  }, [editor, query, caseSensitive, sync])
+
+  // Tear the highlight layer down when the bar closes.
+  useEffect(() => () => clearSearch(editor.state, editor.view.dispatch), [editor])
+
+  // Focus the find field on open.
+  useEffect(() => {
+    findRef.current?.focus()
+    findRef.current?.select()
+  }, [])
+
+  const go = useCallback(
+    (delta: number) => {
+      const st = getSearchState(editor.state)
+      if (!st.matches.length) return
+      setActive(editor.state, editor.view.dispatch, st.active + delta)
+      sync()
+      scrollToActive(editor)
+    },
+    [editor, sync],
+  )
+
+  const replaceActive = useCallback(() => {
+    const st = getSearchState(editor.state)
+    if (st.active < 0) return
+    const m = st.matches[st.active]
+    const tr = editor.state.tr
+    if (replacement) tr.insertText(replacement, m.from, m.to)
+    else tr.delete(m.from, m.to)
+    editor.view.dispatch(tr)
+    sync()
+    scrollToActive(editor)
+  }, [editor, replacement, sync])
+
+  const replaceAll = useCallback(() => {
+    const st = getSearchState(editor.state)
+    if (!st.matches.length) return
+    const tr = editor.state.tr
+    // Apply back-to-front so earlier edits don't shift later match positions.
+    for (let i = st.matches.length - 1; i >= 0; i--) {
+      const m = st.matches[i]
+      if (replacement) tr.insertText(replacement, m.from, m.to)
+      else tr.delete(m.from, m.to)
+    }
+    editor.view.dispatch(tr)
+    sync()
+  }, [editor, replacement, sync])
+
+  const inputStyle: React.CSSProperties = {
+    borderRadius: 'var(--radius-input)',
+    border: '1px solid var(--color-border)',
+    color: 'var(--color-plum)',
+    fontFamily: CJK,
+  }
+
+  return (
+    <div
+      className="petal-no-print absolute right-2 top-2 z-40 flex flex-col gap-1.5 p-2"
+      role="dialog"
+      aria-label="Find and replace"
+      onKeyDown={(e) => {
+        if (e.key === 'Escape') {
+          e.preventDefault()
+          onClose()
+        }
+      }}
+      style={{
+        width: 320,
+        background: 'var(--color-surface)',
+        border: '1px solid var(--color-border)',
+        borderRadius: 'var(--radius-card)',
+        boxShadow: 'var(--shadow-soft)',
+      }}
+    >
+      <div className="flex items-center gap-1.5">
+        <button
+          type="button"
+          onClick={() => setShowReplace((v) => !v)}
+          aria-label={showReplace ? 'Hide replace' : 'Show replace'}
+          title={showReplace ? 'Hide replace' : 'Show replace'}
+          className="flex h-8 w-6 items-center justify-center text-xs"
+          style={{ color: 'var(--color-muted)' }}
+        >
+          {showReplace ? '▾' : '▸'}
+        </button>
+        <input
+          ref={findRef}
+          value={query}
+          onChange={(e) => setQuery(e.target.value)}
+          onKeyDown={(e) => {
+            if (e.key === 'Enter') {
+              e.preventDefault()
+              go(e.shiftKey ? -1 : 1)
+            }
+          }}
+          placeholder="查找 · Find"
+          className="min-w-0 flex-1 px-2 py-1.5 text-sm focus:outline-none"
+          style={inputStyle}
+        />
+        <span
+          className="w-14 shrink-0 text-center text-xs tabular-nums"
+          style={{ color: 'var(--color-muted)' }}
+        >
+          {count ? `${active + 1} / ${count}` : query ? '无 · 0' : ''}
+        </span>
+        <FindBtn label="Previous match" disabled={!count} onClick={() => go(-1)}>↑</FindBtn>
+        <FindBtn label="Next match" disabled={!count} onClick={() => go(1)}>↓</FindBtn>
+        <FindBtn
+          label="Match case"
+          active={caseSensitive}
+          onClick={() => setCaseSensitive((v) => !v)}
+          title="Match case · 区分大小写"
+        >
+          Aa
+        </FindBtn>
+        <FindBtn label="Close" onClick={onClose} title="Close · 关闭">✕</FindBtn>
+      </div>
+
+      {showReplace && (
+        <div className="flex items-center gap-1.5 pl-7">
+          <input
+            value={replacement}
+            onChange={(e) => setReplacement(e.target.value)}
+            onKeyDown={(e) => {
+              if (e.key === 'Enter') {
+                e.preventDefault()
+                replaceActive()
+              }
+            }}
+            placeholder="替换为 · Replace"
+            className="min-w-0 flex-1 px-2 py-1.5 text-sm focus:outline-none"
+            style={inputStyle}
+          />
+          <button
+            type="button"
+            onClick={replaceActive}
+            disabled={!count}
+            className="h-8 shrink-0 whitespace-nowrap px-2.5 text-xs font-semibold disabled:opacity-40"
+            style={{ borderRadius: 'var(--radius-pill)', background: 'var(--color-surface-alt)', color: 'var(--color-plum)' }}
+          >
+            替换
+          </button>
+          <button
+            type="button"
+            onClick={replaceAll}
+            disabled={!count}
+            className="h-8 shrink-0 whitespace-nowrap px-2.5 text-xs font-bold disabled:opacity-40"
+            style={{ borderRadius: 'var(--radius-pill)', background: 'var(--color-accent)', color: '#fff' }}
+          >
+            全部
+          </button>
+        </div>
+      )}
+    </div>
+  )
+}
+
+function FindBtn({
+  children,
+  onClick,
+  label,
+  title,
+  active,
+  disabled,
+}: {
+  children: React.ReactNode
+  onClick: () => void
+  label: string
+  title?: string
+  active?: boolean
+  disabled?: boolean
+}) {
+  return (
+    <button
+      type="button"
+      aria-label={label}
+      title={title ?? label}
+      aria-pressed={active}
+      disabled={disabled}
+      onClick={onClick}
+      className="flex h-8 min-w-8 items-center justify-center px-1.5 text-sm font-semibold disabled:opacity-40"
+      style={{
+        borderRadius: 'var(--radius-input)',
+        color: active ? 'var(--color-accent-hover)' : 'var(--color-muted)',
+        background: active ? 'var(--color-surface-alt)' : 'transparent',
+      }}
+    >
+      {children}
+    </button>
+  )
+}

web/src/components/Editor/FontSize.ts

@@ -0,0 +1,51 @@
+import { Extension } from '@tiptap/core'
+
+// FontSize adds a `fontSize` attribute to the textStyle mark so a writer can
+// pick a size preset (Small / Normal / Large / Title) from the toolbar. It rides
+// on TextStyle (already loaded) rather than introducing a new mark, so it stacks
+// cleanly with color and other inline styling.
+declare module '@tiptap/core' {
+  interface Commands<ReturnType> {
+    fontSize: {
+      setFontSize: (size: string) => ReturnType
+      unsetFontSize: () => ReturnType
+    }
+  }
+}
+
+export const FontSize = Extension.create({
+  name: 'fontSize',
+
+  addOptions() {
+    return { types: ['textStyle'] }
+  },
+
+  addGlobalAttributes() {
+    return [
+      {
+        types: this.options.types,
+        attributes: {
+          fontSize: {
+            default: null,
+            parseHTML: (element) => element.style.fontSize || null,
+            renderHTML: (attributes) =>
+              attributes.fontSize ? { style: `font-size: ${attributes.fontSize}` } : {},
+          },
+        },
+      },
+    ]
+  },
+
+  addCommands() {
+    return {
+      setFontSize:
+        (size) =>
+        ({ chain }) =>
+          chain().setMark('textStyle', { fontSize: size }).run(),
+      unsetFontSize:
+        () =>
+        ({ chain }) =>
+          chain().setMark('textStyle', { fontSize: null }).removeEmptyTextStyle().run(),
+    }
+  },
+})

web/src/components/Editor/SearchHighlight.ts

@@ -0,0 +1,143 @@
+import { Extension } from '@tiptap/core'
+import { Plugin, PluginKey } from '@tiptap/pm/state'
+import type { EditorState, Transaction } from '@tiptap/pm/state'
+import { Decoration, DecorationSet } from '@tiptap/pm/view'
+import type { Node as PMNode } from '@tiptap/pm/model'
+import { mapOffset } from './SuggestionHighlight'
+
+// SearchHighlight powers the in-document Find & Replace bar. Like the suggestion
+// layer it uses ProseMirror *decorations* (not stored marks), so matches are
+// recomputed from the live document on every edit and never leave a stale
+// highlight behind. Matches are found per-textblock (a query never spans a
+// paragraph break), mirroring how the rest of the editor anchors text.
+
+export interface Match {
+  from: number
+  to: number
+}
+
+interface PluginState {
+  query: string
+  caseSensitive: boolean
+  matches: Match[]
+  active: number // index into matches, or -1 when there are none
+  decorations: DecorationSet
+}
+
+export const searchPluginKey = new PluginKey<PluginState>('petalSearch')
+
+// findMatches collects every occurrence of `query` across the document's
+// textblocks and maps each to an absolute ProseMirror range.
+function findMatches(doc: PMNode, query: string, caseSensitive: boolean): Match[] {
+  if (!query) return []
+  const needle = caseSensitive ? query : query.toLowerCase()
+  const matches: Match[] = []
+  doc.descendants((node, pos) => {
+    if (!node.isTextblock) return true
+    const haystackRaw = node.textContent
+    const haystack = caseSensitive ? haystackRaw : haystackRaw.toLowerCase()
+    let idx = haystack.indexOf(needle)
+    while (idx >= 0) {
+      matches.push({
+        from: mapOffset(node, pos, idx),
+        to: mapOffset(node, pos, idx + query.length),
+      })
+      idx = haystack.indexOf(needle, idx + query.length)
+    }
+    return false // don't descend into inline children
+  })
+  return matches
+}
+
+function build(doc: PMNode, query: string, caseSensitive: boolean, preferred: number): PluginState {
+  const matches = findMatches(doc, query, caseSensitive)
+  const active = matches.length === 0 ? -1 : Math.max(0, Math.min(preferred, matches.length - 1))
+  const decos = matches.map((m, i) =>
+    Decoration.inline(m.from, m.to, {
+      class: i === active ? 'petal-find-match petal-find-match-active' : 'petal-find-match',
+    }),
+  )
+  return { query, caseSensitive, matches, active, decorations: DecorationSet.create(doc, decos) }
+}
+
+const EMPTY: PluginState = {
+  query: '',
+  caseSensitive: false,
+  matches: [],
+  active: -1,
+  decorations: DecorationSet.empty,
+}
+
+// setSearch updates the query / case-sensitivity and recomputes matches. Passing
+// an empty query clears the layer.
+export function setSearch(
+  state: EditorState,
+  dispatch: (tr: Transaction) => void,
+  query: string,
+  caseSensitive: boolean,
+) {
+  dispatch(state.tr.setMeta(searchPluginKey, { kind: 'search', query, caseSensitive }))
+}
+
+// setActive moves the highlighted (current) match by index.
+export function setActive(state: EditorState, dispatch: (tr: Transaction) => void, index: number) {
+  dispatch(state.tr.setMeta(searchPluginKey, { kind: 'active', index }))
+}
+
+// clearSearch tears the layer down (on close).
+export function clearSearch(state: EditorState, dispatch: (tr: Transaction) => void) {
+  dispatch(state.tr.setMeta(searchPluginKey, { kind: 'clear' }))
+}
+
+// getSearchState reads the live plugin state (match list + active index) so the
+// Find bar can render "n / m" and drive navigation/replacement.
+export function getSearchState(state: EditorState): PluginState {
+  return searchPluginKey.getState(state) ?? EMPTY
+}
+
+type Meta =
+  | { kind: 'search'; query: string; caseSensitive: boolean }
+  | { kind: 'active'; index: number }
+  | { kind: 'clear' }
+
+export const SearchHighlight = Extension.create({
+  name: 'searchHighlight',
+
+  addProseMirrorPlugins() {
+    return [
+      new Plugin<PluginState>({
+        key: searchPluginKey,
+        state: {
+          init: () => EMPTY,
+          apply(tr, value, _oldState, newState): PluginState {
+            const meta = tr.getMeta(searchPluginKey) as Meta | undefined
+            if (meta?.kind === 'search') {
+              return build(newState.doc, meta.query, meta.caseSensitive, value.active < 0 ? 0 : value.active)
+            }
+            if (meta?.kind === 'active') {
+              if (value.matches.length === 0) return value
+              const active = ((meta.index % value.matches.length) + value.matches.length) % value.matches.length
+              const decos = value.matches.map((m, i) =>
+                Decoration.inline(m.from, m.to, {
+                  class: i === active ? 'petal-find-match petal-find-match-active' : 'petal-find-match',
+                }),
+              )
+              return { ...value, active, decorations: DecorationSet.create(newState.doc, decos) }
+            }
+            if (meta?.kind === 'clear') return EMPTY
+            // Re-anchor on any document change so highlights track edits/replaces.
+            if (tr.docChanged && value.query) {
+              return build(newState.doc, value.query, value.caseSensitive, value.active)
+            }
+            return value
+          },
+        },
+        props: {
+          decorations(state) {
+            return searchPluginKey.getState(state)?.decorations
+          },
+        },
+      }),
+    ]
+  },
+})

web/src/components/Editor/SelectionBubble.tsx

@@ -27,11 +27,14 @@ export const REWRITE_STYLES: RewriteStyle[] = [
 interface Props {
   style: React.CSSProperties
   onRewrite: (style: string) => void
+  // Read the selected text aloud (null when speech isn't available — the button
+  // is then hidden).
+  onSpeak: (() => void) | null
 }
 
 const CJK = "'Nunito','PingFang SC','Microsoft YaHei','Noto Sans CJK SC',sans-serif"
 
-export function SelectionBubble({ style, onRewrite }: Props) {
+export function SelectionBubble({ style, onRewrite, onSpeak }: Props) {
   const [natural, ...tones] = REWRITE_STYLES
 
   return (
@@ -39,21 +42,25 @@ export function SelectionBubble({ style, onRewrite }: Props) {
       className="petal-selection-bubble absolute z-30 flex max-w-[360px] flex-wrap items-center gap-1.5 p-2"
       role="toolbar"
       aria-label="Rewrite the selection"
-      onMouseDown={(e) => e.preventDefault()} // keep the editor selection
       style={{
         background: 'var(--color-surface)',
         border: '1px solid var(--color-border)',
         borderRadius: 'var(--radius-pill)',
         boxShadow: 'var(--shadow-soft)',
         fontFamily: CJK,
+        // Only the buttons capture the pointer; clicks landing on the bubble's
+        // padding/gaps fall through to the text underneath so it never blocks
+        // where the user is trying to click.
+        pointerEvents: 'none',
         ...style,
       }}
     >
       <button
         type="button"
+        onMouseDown={(e) => e.preventDefault()} // keep the editor selection
         onClick={() => onRewrite(natural.value)}
         className="inline-flex h-8 items-center gap-1.5 whitespace-nowrap px-3 text-sm font-bold"
-        style={{ borderRadius: 'var(--radius-pill)', background: 'var(--color-accent)', color: 'var(--color-plum)' }}
+        style={{ borderRadius: 'var(--radius-pill)', background: 'var(--color-accent)', color: 'var(--color-plum)', pointerEvents: 'auto' }}
         title="Rewrite the selection to sound more natural"
       >
         <span aria-hidden>{natural.emoji}</span>
@@ -63,15 +70,30 @@ export function SelectionBubble({ style, onRewrite }: Props) {
         </span>
       </button>
 
+      {onSpeak && (
+        <button
+          type="button"
+          onMouseDown={(e) => e.preventDefault()} // keep the editor selection
+          onClick={onSpeak}
+          className="inline-flex h-8 items-center justify-center px-2 text-sm"
+          style={{ borderRadius: 'var(--radius-pill)', background: 'var(--color-surface-alt)', color: 'var(--color-plum)', pointerEvents: 'auto' }}
+          title="朗读所选 · Read selection aloud"
+          aria-label="Read selection aloud"
+        >
+          🔊
+        </button>
+      )}
+
       <span className="mx-0.5 h-5 w-px shrink-0" style={{ background: 'var(--color-border)' }} />
 
       {tones.map((t) => (
         <button
           key={t.value}
           type="button"
+          onMouseDown={(e) => e.preventDefault()} // keep the editor selection
           onClick={() => onRewrite(t.value)}
           className="inline-flex h-8 items-center gap-1 whitespace-nowrap px-2.5 text-sm font-semibold"
-          style={{ borderRadius: 'var(--radius-pill)', background: 'var(--color-surface-alt)', color: 'var(--color-plum)' }}
+          style={{ borderRadius: 'var(--radius-pill)', background: 'var(--color-surface-alt)', color: 'var(--color-plum)', pointerEvents: 'auto' }}
           onMouseEnter={(e) => (e.currentTarget.style.background = 'var(--color-lavender)')}
           onMouseLeave={(e) => (e.currentTarget.style.background = 'var(--color-surface-alt)')}
           title={`Rewrite in a ${t.en.toLowerCase()} tone`}

web/src/components/Editor/SuggestionHighlight.test.ts

@@ -0,0 +1,54 @@
+import { describe, it, expect } from 'vitest'
+import { foldTypography, findRange } from './SuggestionHighlight'
+import { Schema, type Node as PMNode } from '@tiptap/pm/model'
+
+// A minimal doc/paragraph/text schema, enough to exercise findRange's textblock
+// walk without pulling in the full editor.
+const schema = new Schema({
+  nodes: {
+    doc: { content: 'block+' },
+    paragraph: { group: 'block', content: 'inline*', toDOM: () => ['p', 0] },
+    text: { group: 'inline' },
+  },
+})
+
+// Build a single-paragraph doc with the given plaintext.
+const para = (text: string): PMNode => schema.node('doc', null, [schema.node('paragraph', null, text ? [schema.text(text)] : [])])
+
+describe('foldTypography', () => {
+  it('folds curly quotes back to straight ASCII', () => {
+    expect(foldTypography('“He’s here,” she said').folded).toBe('"He\'s here," she said')
+  })
+
+  it('folds dashes and ellipsis, keeping a source-index map across length changes', () => {
+    const { folded, map } = foldTypography('wait—really...')
+    expect(folded).toBe('wait—really…')
+    // The em dash is already one glyph; the `...` collapsed 3 source chars to 1,
+    // so the trailing sentinel still points just past the last source char.
+    expect(map[map.length - 1]).toBe('wait—really...'.length)
+  })
+
+  it('treats `--` and `...` typed-but-unconverted as their canonical glyphs', () => {
+    expect(foldTypography('a--b...c').folded).toBe('a—b…c')
+  })
+})
+
+describe('findRange typographic anchoring', () => {
+  it('anchors a model echo with straight quotes onto curly-quote document text', () => {
+    const doc = para('She said “hello” to me')
+    const range = findRange(doc, '“hello”'.replace(/[“”]/g, '"')) // model echoed "hello"
+    expect(range).not.toBeNull()
+    // The matched span must cover the curly-quoted region in the real document.
+    expect(doc.textBetween(range!.from - 1, range!.to - 1)).toContain('hello')
+  })
+
+  it('anchors an ASCII `--` echo onto an em-dash in the document', () => {
+    const doc = para('I waited—forever')
+    const range = findRange(doc, 'waited--forever')
+    expect(range).not.toBeNull()
+  })
+
+  it('still returns null when the text genuinely is not present', () => {
+    expect(findRange(para('nothing to see'), 'absent phrase')).toBeNull()
+  })
+})

web/src/components/Editor/SuggestionHighlight.ts

@@ -42,21 +42,87 @@ export function mapOffset(block: PMNode, blockPos: number, targetOffset: number)
   return result
 }
 
+// foldTypography canonicalizes the typographic variants that the editor's
+// Typography input rules introduce (curly quotes, em/en dashes, single-glyph
+// ellipsis) and the unicode spaces a paste can carry. The model echoes a
+// suggestion's `original` from the same text but often normalizes these back to
+// plain ASCII — straight quotes, `--`, `...` — so an exact match would miss and
+// the suggestion could be neither highlighted nor applied.
+//
+// It returns the folded string plus `map`, where map[k] is the source index at
+// which folded character k begins. A multi-character source run (`...`, `--`)
+// folds to one glyph, so lengths differ; map projects a match in folded space
+// back onto real document offsets. map has a trailing sentinel (= source length)
+// so map[start + folded.length] yields the offset just past the match.
+export function foldTypography(text: string): { folded: string; map: number[] } {
+  let folded = ''
+  const map: number[] = []
+  let i = 0
+  while (i < text.length) {
+    // Multi-character ASCII sequences the Typography rules would have replaced.
+    if (text.startsWith('...', i)) {
+      folded += '…'
+      map.push(i)
+      i += 3
+      continue
+    }
+    if (text[i] === '-' && text[i + 1] === '-') {
+      folded += '—'
+      map.push(i)
+      i += 2
+      continue
+    }
+    folded += foldChar(text[i])
+    map.push(i)
+    i++
+  }
+  map.push(text.length)
+  return { folded, map }
+}
+
+// foldChar maps a single character onto its canonical form (length-preserving).
+function foldChar(c: string): string {
+  switch (c) {
+    case '‘':
+    case '’':
+    case '‚':
+    case '‛':
+      return "'"
+    case '“':
+    case '”':
+    case '„':
+    case '‟':
+      return '"'
+    case '–': // en dash → em dash, the canonical form `--` also folds to
+      return '—'
+    case ' ': // non-breaking space
+    case ' ': // thin space
+    case ' ': // narrow no-break space
+      return ' '
+    default:
+      return c
+  }
+}
+
 // findRange locates the first occurrence of `search` within a single textblock
 // and returns its ProseMirror range, or null if the string isn't present (the
-// user may have edited or removed it since the checkpoint ran). Exported so the
+// user may have edited or removed it since the checkpoint ran). Matching is done
-// accept flow can resolve the same span to replace.
+// in canonical typographic space (see foldTypography) so curly-quote/dash/ellipsis
+// differences between the model's echo and the live document don't strand the
+// anchor. Exported so the accept flow can resolve the same span to replace.
 export function findRange(doc: PMNode, search: string): { from: number; to: number } | null {
-  if (!search) return null
+  const needle = foldTypography(search).folded
+  if (!needle) return null
   let result: { from: number; to: number } | null = null
   doc.descendants((node, pos) => {
     if (result) return false
     if (!node.isTextblock) return true // keep descending to the textblock
-    const idx = node.textContent.indexOf(search)
+    const { folded, map } = foldTypography(node.textContent)
+    const idx = folded.indexOf(needle)
     if (idx >= 0) {
       result = {
-        from: mapOffset(node, pos, idx),
+        from: mapOffset(node, pos, map[idx]),
-        to: mapOffset(node, pos, idx + search.length),
+        to: mapOffset(node, pos, map[idx + needle.length]),
       }
     }
     return false // never descend into a textblock's inline children

web/src/components/Editor/Typography.ts

@@ -0,0 +1,29 @@
+import { Extension, textInputRule } from '@tiptap/core'
+
+// Typography adds the small print-shop niceties as you type: curly quotes,
+// em-dashes, and a single-character ellipsis. Implemented as input rules (no
+// dependency, offline-friendly — matching FontSize.ts's hand-rolled approach),
+// and every one is plain Undo-able if it ever fires when you didn't want it.
+//
+// CJK is untouched: these rules only rewrite ASCII quotes/dashes/dots, never the
+// fullwidth forms a Mandarin sentence uses.
+export const Typography = Extension.create({
+  name: 'petalTypography',
+
+  addInputRules() {
+    return [
+      // “ opening double quote — after start, whitespace, or an opening bracket.
+      textInputRule({ find: /(?:^|[\s{[(<'"‘“])(")$/, replace: '“' }),
+      // ” closing double quote — any other time you type a ".
+      textInputRule({ find: /"$/, replace: '”' }),
+      // ‘ opening single quote.
+      textInputRule({ find: /(?:^|[\s{[(<'"‘“])(')$/, replace: '‘' }),
+      // ’ closing single quote / apostrophe.
+      textInputRule({ find: /'$/, replace: '’' }),
+      // — em dash from a double hyphen.
+      textInputRule({ find: /--$/, replace: '—' }),
+      // … ellipsis from three dots.
+      textInputRule({ find: /\.\.\.$/, replace: '…' }),
+    ]
+  },
+})

web/src/components/Editor/WordCard.tsx

@@ -1,4 +1,5 @@
 import type { WordInfo } from '../../api/client'
+import { speak, speechSupported } from '../../audio/speech'
 
 // WordCard is the right-click popover for any word: its dictionary definition(s)
 // on top and tappable synonym pills below. Clicking a synonym replaces the word
@@ -18,6 +19,7 @@ export function WordCard({ word, info, loading, style, onReplace }: Props) {
   const definitions = info?.definitions ?? []
   const synonyms = info?.synonyms ?? []
   const gloss = info?.gloss ?? ''
+  const phonetic = info?.phonetic ?? ''
   const empty = !loading && !gloss && definitions.length === 0 && synonyms.length === 0
 
   return (
@@ -46,8 +48,28 @@ export function WordCard({ word, info, loading, style, onReplace }: Props) {
         <span className="font-bold" style={{ color: 'var(--color-plum)' }}>
           {word}
         </span>
+        {speechSupported() && (
+          <button
+            type="button"
+            onClick={() => speak(word)}
+            aria-label={`Pronounce ${word}`}
+            title="朗读 · Read aloud"
+            className="ml-auto flex h-7 w-7 items-center justify-center rounded-full text-sm"
+            style={{ background: 'var(--color-surface-alt)', color: 'var(--color-plum)' }}
+          >
+            🔊
+          </button>
+        )}
       </div>
 
+      {/* How to say it — the pronunciation aid for an English learner, paired
+          with the 🔊 button above. */}
+      {phonetic && (
+        <p className="mt-1.5 text-sm" style={{ color: 'var(--color-muted)' }}>
+          /{phonetic}/
+        </p>
+      )}
+
       {/* Chinese gloss first — it's what the Mandarin-speaking writer reaches for. */}
       {gloss && (
         <p

web/src/components/StatusBar/SoundToggle.tsx

@@ -0,0 +1,37 @@
+import { useEffect, useState } from 'react'
+import { isSoundEnabled, onSoundEnabledChange, playPop, setSoundEnabled } from '../../audio/sounds'
+
+// A tiny mute toggle for Petal's cute sounds, tucked at the right of the status
+// bar. Bilingual tooltip (she reads Mandarin first), and a soft confirming pop
+// when sounds are turned back on so the choice is audible.
+export function SoundToggle() {
+  const [on, setOn] = useState(isSoundEnabled)
+
+  // Stay in sync if the setting is flipped elsewhere.
+  useEffect(() => onSoundEnabledChange(setOn), [])
+
+  const toggle = () => {
+    const next = !on
+    setSoundEnabled(next)
+    setOn(next)
+    if (next) playPop() // a little hello when re-enabled
+  }
+
+  return (
+    <button
+      type="button"
+      onClick={toggle}
+      aria-pressed={on}
+      title={on ? '声音开 · Sounds on' : '声音关 · Sounds off'}
+      aria-label={on ? 'Mute Petal sounds' : 'Unmute Petal sounds'}
+      className="flex items-center justify-center rounded-full px-2 py-1 text-xl leading-none transition-colors"
+      style={{ color: on ? 'var(--color-accent)' : 'var(--color-muted)', lineHeight: 1 }}
+      onMouseEnter={(e) => (e.currentTarget.style.color = 'var(--color-accent-hover)')}
+      onMouseLeave={(e) =>
+        (e.currentTarget.style.color = on ? 'var(--color-accent)' : 'var(--color-muted)')
+      }
+    >
+      <span aria-hidden>{on ? '🔔' : '🔕'}</span>
+    </button>
+  )
+}

web/src/components/StatusBar/StatusBar.tsx

@@ -1,6 +1,7 @@
 import { useEffect, useRef, useState } from 'react'
 import type { SaveStatus } from '../../hooks/useAutoSave'
 import { StatsPanel } from './StatsPanel'
+import { SoundToggle } from './SoundToggle'
 
 interface Props {
   wordCount: number
@@ -44,7 +45,7 @@ export function StatusBar({ wordCount, text, saveStatus, checking, voicing, llmD
 
   return (
     <footer
-      className="flex h-9 shrink-0 items-center gap-3 px-6 text-xs"
+      className="flex h-11 shrink-0 items-center gap-3 px-6 text-sm"
       style={{ borderTop: '1px solid var(--color-border)', color: 'var(--color-muted)' }}
     >
       <div className="relative" ref={statsRef}>
@@ -53,7 +54,7 @@ export function StatusBar({ wordCount, text, saveStatus, checking, voicing, llmD
           onClick={() => setStatsOpen((o) => !o)}
           aria-haspopup="dialog"
           aria-expanded={statsOpen}
-          className="rounded-full px-1.5 py-0.5 font-semibold transition-colors"
+          className="rounded-full px-2 py-0.5 text-[0.95rem] font-bold transition-colors"
           style={{ color: statsOpen ? 'var(--color-accent-hover)' : 'inherit' }}
           onMouseEnter={(e) => (e.currentTarget.style.color = 'var(--color-accent-hover)')}
           onMouseLeave={(e) =>
@@ -120,6 +121,9 @@ export function StatusBar({ wordCount, text, saveStatus, checking, voicing, llmD
           </span>
         </>
       )}
+      <div className="ml-auto">
+        <SoundToggle />
+      </div>
     </footer>
   )
 }

web/src/components/Toolbar/Toolbar.tsx

@@ -1,5 +1,7 @@
 import type { Editor } from '@tiptap/react'
 import { useEditorState } from '@tiptap/react'
+import { useEffect, useRef, useState } from 'react'
+import { uploadImageInto } from '../Editor/EditorCore'
 
 interface Props {
   editor: Editor | null
@@ -15,18 +17,21 @@ function TBtn({
   active,
   disabled,
   label,
+  title,
   children,
 }: {
   onClick: () => void
   active?: boolean
   disabled?: boolean
   label: string
+  title?: string
   children: React.ReactNode
 }) {
   return (
     <button
       type="button"
       aria-label={label}
+      title={title ?? label}
       aria-pressed={active}
       disabled={disabled}
       onMouseDown={(e) => e.preventDefault()} // keep editor selection
@@ -47,10 +52,125 @@ const Divider = () => (
   <span className="mx-1 h-5 w-px" style={{ background: 'var(--color-border)' }} />
 )
 
+// A popover anchored under its trigger. The trigger + panel share a relative
+// wrapper; `open`/`onClose` are owned by the toolbar so only one is open at once.
+// A pointer-down outside the wrapper closes it.
+function Popover({
+  open,
+  onClose,
+  trigger,
+  children,
+  width = 200,
+}: {
+  open: boolean
+  onClose: () => void
+  trigger: React.ReactNode
+  children: React.ReactNode
+  width?: number
+}) {
+  const ref = useRef<HTMLDivElement>(null)
+  useEffect(() => {
+    if (!open) return
+    const onDown = (e: MouseEvent) => {
+      if (!ref.current?.contains(e.target as Node)) onClose()
+    }
+    document.addEventListener('mousedown', onDown)
+    return () => document.removeEventListener('mousedown', onDown)
+  }, [open, onClose])
+  return (
+    <div ref={ref} className="relative flex items-center">
+      {trigger}
+      {open && (
+        <div
+          className="absolute left-0 top-full z-40 mt-1.5 p-2"
+          style={{
+            width,
+            borderRadius: 'var(--radius-card)',
+            background: 'var(--color-surface)',
+            border: '1px solid var(--color-border)',
+            boxShadow: 'var(--shadow-soft)',
+          }}
+        >
+          {children}
+        </div>
+      )}
+    </div>
+  )
+}
+
+// Text-color swatches. `null` clears the color back to the theme default.
+const TEXT_COLORS: { label: string; value: string | null }[] = [
+  { label: 'Default', value: null },
+  { label: 'Rose', value: '#D98AAF' },
+  { label: 'Coral', value: '#E0785C' },
+  { label: 'Honey', value: '#C68B2E' },
+  { label: 'Green', value: '#4F9E7F' },
+  { label: 'Sky', value: '#4E8FCC' },
+  { label: 'Lavender', value: '#7D63C4' },
+  { label: 'Plum', value: '#3D2E39' },
+]
+
+// Highlighter colors — soft pastels so dark text stays readable on top.
+const HIGHLIGHTS: { label: string; value: string | null }[] = [
+  { label: 'None', value: null },
+  { label: 'Yellow', value: '#FFF1A8' },
+  { label: 'Pink', value: '#FAD4E4' },
+  { label: 'Mint', value: '#CDEFE2' },
+  { label: 'Peach', value: '#FBE0CF' },
+  { label: 'Lavender', value: '#E6DCFA' },
+  { label: 'Sky', value: '#D6E8FB' },
+]
+
+// Font-size presets. `null` clears back to the document default.
+const SIZES: { label: string; value: string | null; em: string }[] = [
+  { label: 'Small', value: '0.85em', em: '0.85em' },
+  { label: 'Normal', value: null, em: '1em' },
+  { label: 'Large', value: '1.3em', em: '1.3em' },
+  { label: 'Title', value: '1.7em', em: '1.7em' },
+]
+
+// A round color chip used in the color/highlight palettes.
+function Swatch({
+  color,
+  active,
+  onClick,
+  title,
+}: {
+  color: string | null
+  active: boolean
+  onClick: () => void
+  title: string
+}) {
+  return (
+    <button
+      type="button"
+      title={title}
+      aria-label={title}
+      onMouseDown={(e) => e.preventDefault()}
+      onClick={onClick}
+      className="flex h-7 w-7 items-center justify-center"
+      style={{
+        borderRadius: 'var(--radius-pill)',
+        background: color ?? 'var(--color-surface)',
+        border: active ? '2px solid var(--color-accent)' : '1px solid var(--color-border)',
+        // The "clear" chip (no color) shows a tiny diagonal stroke.
+        backgroundImage: color
+          ? undefined
+          : 'linear-gradient(135deg, transparent 44%, var(--color-accent) 44%, var(--color-accent) 56%, transparent 56%)',
+      }}
+    />
+  )
+}
+
 // Toolbar renders inline formatting controls bound to the live Tiptap editor.
 // useEditorState subscribes to just the flags it reads, so the buttons reflect
 // the current selection without re-rendering the whole tree on every keystroke.
 export function Toolbar({ editor, onVoiceCheck, voicing }: Props) {
+  // Which popover (if any) is open. Only one at a time.
+  const [menu, setMenu] = useState<'color' | 'highlight' | 'size' | 'link' | 'table' | 'outline' | null>(null)
+  const [linkUrl, setLinkUrl] = useState('')
+  const fileInputRef = useRef<HTMLInputElement>(null)
+
   const state = useEditorState({
     editor,
     selector: ({ editor }) =>
@@ -59,27 +179,92 @@ export function Toolbar({ editor, onVoiceCheck, voicing }: Props) {
             bold: editor.isActive('bold'),
             italic: editor.isActive('italic'),
             underline: editor.isActive('underline'),
+            strike: editor.isActive('strike'),
             h1: editor.isActive('heading', { level: 1 }),
             h2: editor.isActive('heading', { level: 2 }),
+            h3: editor.isActive('heading', { level: 3 }),
             bullet: editor.isActive('bulletList'),
+            ordered: editor.isActive('orderedList'),
             left: editor.isActive({ textAlign: 'left' }),
             center: editor.isActive({ textAlign: 'center' }),
             right: editor.isActive({ textAlign: 'right' }),
+            link: editor.isActive('link'),
+            inTable: editor.isActive('table'),
+            color: editor.getAttributes('textStyle').color ?? null,
+            highlight: editor.getAttributes('highlight').color ?? null,
+            fontSize: editor.getAttributes('textStyle').fontSize ?? null,
+            canUndo: editor.can().undo(),
+            canRedo: editor.can().redo(),
           }
         : null,
   })
 
   if (!editor || !state) return null
 
+  const close = () => setMenu(null)
+
+  // Open the link popover, pre-filling the field with any link already on the
+  // selection so it can be edited rather than retyped.
+  const openLink = () => {
+    setLinkUrl(editor.getAttributes('link').href ?? '')
+    setMenu(menu === 'link' ? null : 'link')
+  }
+
+  const applyLink = () => {
+    const url = linkUrl.trim()
+    if (!url) {
+      editor.chain().focus().extendMarkRange('link').unsetLink().run()
+    } else {
+      // Default to https:// when the writer omits a scheme.
+      const href = /^(https?:|mailto:|\/)/i.test(url) ? url : `https://${url}`
+      editor.chain().focus().extendMarkRange('link').setLink({ href }).run()
+    }
+    close()
+  }
+
+  // Collect the document's headings (with their positions) for the outline. Read
+  // fresh each time the popover opens, so it always reflects the current doc.
+  const headings: { level: number; text: string; pos: number }[] = []
+  if (menu === 'outline') {
+    editor.state.doc.descendants((node, pos) => {
+      if (node.type.name === 'heading') {
+        headings.push({ level: (node.attrs.level as number) || 1, text: node.textContent || '（无标题）', pos })
+      }
+    })
+  }
+
+  // Scroll a heading into view without moving the selection (which would pop the
+  // rewrite bubble). domAtPos resolves the heading's DOM node to scroll to.
+  const gotoHeading = (pos: number) => {
+    const dom = editor.view.domAtPos(pos + 1)
+    const el = dom.node.nodeType === Node.TEXT_NODE ? dom.node.parentElement : (dom.node as HTMLElement)
+    el?.scrollIntoView({ block: 'start', behavior: 'smooth' })
+    close()
+  }
+
+  const onPickImage = (e: React.ChangeEvent<HTMLInputElement>) => {
+    const file = e.target.files?.[0]
+    if (file) uploadImageInto(editor.view, file)
+    e.target.value = '' // allow re-picking the same file
+  }
+
   return (
     <div
-      className="mb-4 flex items-center gap-0.5 self-start px-2 py-1.5"
+      className="petal-toolbar mb-4 flex items-center gap-0.5 self-start px-2 py-1.5"
       style={{
-        borderRadius: 'var(--radius-pill)',
+        borderRadius: 'var(--radius-card)',
         background: 'var(--color-surface)',
         boxShadow: 'var(--shadow-soft)',
       }}
     >
+      <TBtn label="Undo" disabled={!state.canUndo} onClick={() => editor.chain().focus().undo().run()}>
+        ↶
+      </TBtn>
+      <TBtn label="Redo" disabled={!state.canRedo} onClick={() => editor.chain().focus().redo().run()}>
+        ↷
+      </TBtn>
+      <Divider />
+
       <TBtn label="Bold" active={state.bold} onClick={() => editor.chain().focus().toggleBold().run()}>
         <span className="font-extrabold">B</span>
       </TBtn>
@@ -93,51 +278,301 @@ export function Toolbar({ editor, onVoiceCheck, voicing }: Props) {
       >
         <span className="underline">U</span>
       </TBtn>
+      <TBtn label="Strikethrough" active={state.strike} onClick={() => editor.chain().focus().toggleStrike().run()}>
+        <span className="line-through">S</span>
+      </TBtn>
       <Divider />
-      <TBtn
+
-        label="Heading 1"
+      {/* Text color */}
-        active={state.h1}
+      <Popover
-        onClick={() => editor.chain().focus().toggleHeading({ level: 1 }).run()}
+        open={menu === 'color'}
+        onClose={close}
+        width={188}
+        trigger={
+          <TBtn label="Text color" active={menu === 'color'} onClick={() => setMenu(menu === 'color' ? null : 'color')}>
+            <span className="flex flex-col items-center leading-none">
+              <span className="text-sm font-bold">A</span>
+              <span
+                className="mt-0.5 h-1 w-4 rounded-full"
+                style={{ background: state.color ?? 'var(--color-accent)' }}
+              />
+            </span>
+          </TBtn>
+        }
       >
+        <div className="grid grid-cols-4 gap-1.5">
+          {TEXT_COLORS.map((c) => (
+            <Swatch
+              key={c.label}
+              color={c.value}
+              title={c.label}
+              active={state.color === c.value}
+              onClick={() => {
+                if (c.value) editor.chain().focus().setColor(c.value).run()
+                else editor.chain().focus().unsetColor().run()
+                close()
+              }}
+            />
+          ))}
+        </div>
+      </Popover>
+
+      {/* Highlight */}
+      <Popover
+        open={menu === 'highlight'}
+        onClose={close}
+        width={170}
+        trigger={
+          <TBtn
+            label="Highlight"
+            active={menu === 'highlight' || !!state.highlight}
+            onClick={() => setMenu(menu === 'highlight' ? null : 'highlight')}
+          >
+            <span
+              className="flex h-5 w-5 items-center justify-center rounded text-xs font-bold"
+              style={{ background: state.highlight ?? '#FFF1A8', color: 'var(--color-plum)' }}
+            >
+              H
+            </span>
+          </TBtn>
+        }
+      >
+        <div className="grid grid-cols-4 gap-1.5">
+          {HIGHLIGHTS.map((c) => (
+            <Swatch
+              key={c.label}
+              color={c.value}
+              title={c.label}
+              active={state.highlight === c.value}
+              onClick={() => {
+                if (c.value) editor.chain().focus().setHighlight({ color: c.value }).run()
+                else editor.chain().focus().unsetHighlight().run()
+                close()
+              }}
+            />
+          ))}
+        </div>
+      </Popover>
+
+      {/* Font size */}
+      <Popover
+        open={menu === 'size'}
+        onClose={close}
+        width={140}
+        trigger={
+          <TBtn label="Text size" active={menu === 'size'} onClick={() => setMenu(menu === 'size' ? null : 'size')}>
+            <span className="text-xs font-bold tracking-tight">
+              <span className="text-[0.7rem]">A</span>
+              <span className="text-sm">A</span>
+            </span>
+          </TBtn>
+        }
+      >
+        <div className="flex flex-col gap-0.5">
+          {SIZES.map((s) => {
+            const active = state.fontSize === s.value
+            return (
+              <button
+                key={s.label}
+                type="button"
+                onMouseDown={(e) => e.preventDefault()}
+                onClick={() => {
+                  if (s.value) editor.chain().focus().setFontSize(s.value).run()
+                  else editor.chain().focus().unsetFontSize().run()
+                  close()
+                }}
+                className="flex items-center justify-between rounded px-2 py-1 text-left"
+                style={{
+                  background: active ? 'var(--color-surface-alt)' : 'transparent',
+                  color: active ? 'var(--color-accent-hover)' : 'var(--color-plum)',
+                }}
+              >
+                <span style={{ fontSize: s.em }}>{s.label}</span>
+              </button>
+            )
+          })}
+        </div>
+      </Popover>
+      <Divider />
+
+      <TBtn label="Heading 1" active={state.h1} onClick={() => editor.chain().focus().toggleHeading({ level: 1 }).run()}>
         H1
       </TBtn>
-      <TBtn
+      <TBtn label="Heading 2" active={state.h2} onClick={() => editor.chain().focus().toggleHeading({ level: 2 }).run()}>
-        label="Heading 2"
-        active={state.h2}
-        onClick={() => editor.chain().focus().toggleHeading({ level: 2 }).run()}
-      >
         H2
       </TBtn>
-      <TBtn
+      <TBtn label="Heading 3" active={state.h3} onClick={() => editor.chain().focus().toggleHeading({ level: 3 }).run()}>
-        label="Bullet list"
+        H3
-        active={state.bullet}
+      </TBtn>
-        onClick={() => editor.chain().focus().toggleBulletList().run()}
+      <TBtn label="Bullet list" active={state.bullet} onClick={() => editor.chain().focus().toggleBulletList().run()}>
-      >
         •
       </TBtn>
+      <TBtn label="Numbered list" active={state.ordered} onClick={() => editor.chain().focus().toggleOrderedList().run()}>
+        1.
+      </TBtn>
       <Divider />
-      <TBtn
+
-        label="Align left"
+      <TBtn label="Align left" active={state.left} onClick={() => editor.chain().focus().setTextAlign('left').run()}>
-        active={state.left}
-        onClick={() => editor.chain().focus().setTextAlign('left').run()}
-      >
         ⇤
       </TBtn>
-      <TBtn
+      <TBtn label="Align center" active={state.center} onClick={() => editor.chain().focus().setTextAlign('center').run()}>
-        label="Align center"
-        active={state.center}
-        onClick={() => editor.chain().focus().setTextAlign('center').run()}
-      >
         ↔
       </TBtn>
-      <TBtn
+      <TBtn label="Align right" active={state.right} onClick={() => editor.chain().focus().setTextAlign('right').run()}>
-        label="Align right"
-        active={state.right}
-        onClick={() => editor.chain().focus().setTextAlign('right').run()}
-      >
         ⇥
       </TBtn>
       <Divider />
+
+      {/* Link */}
+      <Popover
+        open={menu === 'link'}
+        onClose={close}
+        width={236}
+        trigger={
+          <TBtn label="Link" active={state.link || menu === 'link'} onClick={openLink}>
+            🔗
+          </TBtn>
+        }
+      >
+        <div className="flex flex-col gap-2">
+          <input
+            autoFocus
+            value={linkUrl}
+            onChange={(e) => setLinkUrl(e.target.value)}
+            onKeyDown={(e) => {
+              if (e.key === 'Enter') applyLink()
+              if (e.key === 'Escape') close()
+            }}
+            placeholder="https://…"
+            className="w-full px-2 py-1.5 text-sm focus:outline-none"
+            style={{
+              borderRadius: 'var(--radius-input)',
+              border: '1px solid var(--color-border)',
+              color: 'var(--color-plum)',
+            }}
+          />
+          <div className="flex items-center justify-between">
+            {state.link ? (
+              <button
+                type="button"
+                onMouseDown={(e) => e.preventDefault()}
+                onClick={() => {
+                  editor.chain().focus().extendMarkRange('link').unsetLink().run()
+                  close()
+                }}
+                className="px-2 py-1 text-xs font-semibold"
+                style={{ color: 'var(--color-muted)' }}
+              >
+                Remove
+              </button>
+            ) : (
+              <span />
+            )}
+            <button
+              type="button"
+              onMouseDown={(e) => e.preventDefault()}
+              onClick={applyLink}
+              className="px-3 py-1 text-xs font-bold"
+              style={{
+                borderRadius: 'var(--radius-pill)',
+                background: 'var(--color-accent)',
+                color: '#fff',
+              }}
+            >
+              Apply
+            </button>
+          </div>
+        </div>
+      </Popover>
+
+      {/* Image */}
+      <input ref={fileInputRef} type="file" accept="image/*" hidden onChange={onPickImage} />
+      <TBtn label="Insert image" onClick={() => fileInputRef.current?.click()}>
+        🖼️
+      </TBtn>
+
+      {/* Table */}
+      <Popover
+        open={menu === 'table'}
+        onClose={close}
+        width={state.inTable ? 188 : 150}
+        trigger={
+          <TBtn label="Table" active={state.inTable || menu === 'table'} onClick={() => setMenu(menu === 'table' ? null : 'table')}>
+            ▦
+          </TBtn>
+        }
+      >
+        {state.inTable ? (
+          <div className="flex flex-col gap-0.5 text-sm" style={{ color: 'var(--color-plum)' }}>
+            <TableAction label="Add row below" onClick={() => editor.chain().focus().addRowAfter().run()} />
+            <TableAction label="Add column right" onClick={() => editor.chain().focus().addColumnAfter().run()} />
+            <TableAction label="Toggle header row" onClick={() => editor.chain().focus().toggleHeaderRow().run()} />
+            <TableAction label="Delete row" onClick={() => editor.chain().focus().deleteRow().run()} />
+            <TableAction label="Delete column" onClick={() => editor.chain().focus().deleteColumn().run()} />
+            <TableAction
+              label="Delete table"
+              danger
+              onClick={() => {
+                editor.chain().focus().deleteTable().run()
+                close()
+              }}
+            />
+          </div>
+        ) : (
+          <GridPicker
+            onPick={(rows, cols) => {
+              editor.chain().focus().insertTable({ rows, cols, withHeaderRow: true }).run()
+              close()
+            }}
+          />
+        )}
+      </Popover>
+
+      {/* Outline / document map */}
+      <Popover
+        open={menu === 'outline'}
+        onClose={close}
+        width={240}
+        trigger={
+          <TBtn label="Outline" active={menu === 'outline'} onClick={() => setMenu(menu === 'outline' ? null : 'outline')}>
+            ☰
+          </TBtn>
+        }
+      >
+        <div className="max-h-72 overflow-y-auto">
+          <p className="mb-1.5 px-1 text-xs font-bold" style={{ color: 'var(--color-muted)' }}>
+            大纲 · Outline
+          </p>
+          {headings.length === 0 ? (
+            <p className="px-1 py-2 text-xs leading-snug" style={{ color: 'var(--color-muted)' }}>
+              用 H1/H2/H3 添加标题，这里就会出现导航。<br />
+              Add headings to navigate them here.
+            </p>
+          ) : (
+            <div className="flex flex-col">
+              {headings.map((h, i) => (
+                <button
+                  key={i}
+                  type="button"
+                  onMouseDown={(e) => e.preventDefault()}
+                  onClick={() => gotoHeading(h.pos)}
+                  className="truncate rounded px-1.5 py-1 text-left text-sm hover:bg-[var(--color-surface-alt)]"
+                  style={{
+                    paddingLeft: `${(h.level - 1) * 12 + 6}px`,
+                    color: 'var(--color-plum)',
+                    fontWeight: h.level === 1 ? 700 : 500,
+                  }}
+                  title={h.text}
+                >
+                  {h.text}
+                </button>
+              ))}
+            </div>
+          )}
+        </div>
+      </Popover>
+      <Divider />
+
       <button
         type="button"
         aria-label="Check my voice"
@@ -162,3 +597,58 @@ export function Toolbar({ editor, onVoiceCheck, voicing }: Props) {
     </div>
   )
 }
+
+// A row in the in-table editing menu.
+function TableAction({ label, onClick, danger }: { label: string; onClick: () => void; danger?: boolean }) {
+  return (
+    <button
+      type="button"
+      onMouseDown={(e) => e.preventDefault()}
+      onClick={onClick}
+      className="rounded px-2 py-1 text-left"
+      style={{ color: danger ? 'var(--color-accent-hover)' : 'var(--color-plum)' }}
+    >
+      {label}
+    </button>
+  )
+}
+
+// A hover-to-size grid for inserting a table. Hovering a cell highlights the
+// rectangle from the top-left; clicking inserts that many rows × columns.
+function GridPicker({ onPick }: { onPick: (rows: number, cols: number) => void }) {
+  const MAX = 6
+  const [hover, setHover] = useState<{ r: number; c: number }>({ r: 0, c: 0 })
+  return (
+    <div>
+      <div
+        className="grid gap-0.5"
+        style={{ gridTemplateColumns: `repeat(${MAX}, 1fr)` }}
+        onMouseLeave={() => setHover({ r: 0, c: 0 })}
+      >
+        {Array.from({ length: MAX * MAX }).map((_, i) => {
+          const r = Math.floor(i / MAX) + 1
+          const c = (i % MAX) + 1
+          const on = r <= hover.r && c <= hover.c
+          return (
+            <button
+              key={i}
+              type="button"
+              onMouseDown={(e) => e.preventDefault()}
+              onMouseEnter={() => setHover({ r, c })}
+              onClick={() => onPick(r, c)}
+              className="h-4 w-4"
+              style={{
+                borderRadius: 3,
+                background: on ? 'var(--color-accent)' : 'var(--color-surface-alt)',
+                border: '1px solid var(--color-border)',
+              }}
+            />
+          )
+        })}
+      </div>
+      <div className="mt-1.5 text-center text-xs" style={{ color: 'var(--color-muted)' }}>
+        {hover.r > 0 ? `${hover.r} × ${hover.c}` : 'Pick a size'}
+      </div>
+    </div>
+  )
+}

web/src/effects/PetalFall.tsx

@@ -0,0 +1,200 @@
+import { useEffect, useRef } from 'react'
+
+// PetalFall is the cozy ambient layer: a fixed, non-interactive 2D canvas that
+// drifts soft cherry-blossom petals down the page. Each petal is a real
+// petal-shaped sprite (notched sakura silhouette with a gentle gradient and
+// soft alpha), pre-rendered once per color and then blitted many times with
+// rotation + sway. It sits behind the chrome, ignores pointer events, pauses
+// when the tab is hidden, and renders nothing for prefers-reduced-motion.
+
+// Saturated-but-soft blossom colors. Each petal fades from a light, near-white
+// tip to its color at the base, so it reads as a real petal rather than a blob.
+const PALETTE: { r: number; g: number; b: number }[] = [
+  { r: 255, g: 150, b: 190 }, // rose pink
+  { r: 255, g: 184, b: 205 }, // pale blossom
+  { r: 255, g: 158, b: 130 }, // warm peach
+  { r: 209, g: 160, b: 255 }, // lavender
+  { r: 255, g: 122, b: 172 }, // deep rose
+]
+
+const SPRITE_PX = 160 // off-screen render resolution per petal sprite
+
+function rgba(c: { r: number; g: number; b: number }, a: number): string {
+  return `rgba(${Math.round(c.r)}, ${Math.round(c.g)}, ${Math.round(c.b)}, ${a})`
+}
+function mixWhite(c: { r: number; g: number; b: number }, t: number) {
+  return { r: c.r + (255 - c.r) * t, g: c.g + (255 - c.g) * t, b: c.b + (255 - c.b) * t }
+}
+function scale(c: { r: number; g: number; b: number }, t: number) {
+  return { r: c.r * t, g: c.g * t, b: c.b * t }
+}
+
+// Draw one petal sprite: a notched sakura silhouette filled with a tip→base
+// gradient and a faint central highlight, on a transparent canvas.
+function makeSprite(color: { r: number; g: number; b: number }): HTMLCanvasElement {
+  const s = SPRITE_PX
+  const cv = document.createElement('canvas')
+  cv.width = s
+  cv.height = s
+  const g = cv.getContext('2d')!
+  g.translate(s / 2, s / 2)
+
+  const hw = s * 0.3 // half width
+  const hh = s * 0.44 // half height
+
+  g.beginPath()
+  g.moveTo(0, hh) // pointed base
+  g.bezierCurveTo(hw, hh * 0.42, hw * 0.92, -hh * 0.52, hw * 0.2, -hh * 0.92) // right side → near tip
+  g.quadraticCurveTo(0, -hh * 0.72, -hw * 0.2, -hh * 0.92) // the sakura notch
+  g.bezierCurveTo(-hw * 0.92, -hh * 0.52, -hw, hh * 0.42, 0, hh) // left side → base
+  g.closePath()
+
+  const grad = g.createLinearGradient(0, -hh, 0, hh)
+  grad.addColorStop(0, rgba(mixWhite(color, 0.55), 0.78)) // light, soft tip
+  grad.addColorStop(0.45, rgba(color, 0.88))
+  grad.addColorStop(1, rgba(scale(color, 0.8), 0.95)) // deeper base
+  g.fillStyle = grad
+  g.fill()
+
+  // A soft white sheen down the centre for a delicate, glossy feel.
+  const sheen = g.createLinearGradient(-hw * 0.3, 0, hw * 0.3, 0)
+  sheen.addColorStop(0, 'rgba(255,255,255,0)')
+  sheen.addColorStop(0.5, 'rgba(255,255,255,0.28)')
+  sheen.addColorStop(1, 'rgba(255,255,255,0)')
+  g.fillStyle = sheen
+  g.fill()
+
+  return cv
+}
+
+interface Petal {
+  baseX: number
+  y: number
+  size: number
+  vy: number // fall speed (px/s)
+  angle: number
+  spin: number // rad/s
+  swayAmp: number
+  swayFreq: number
+  swayPhase: number
+  alpha: number
+  sprite: number
+}
+
+export function PetalFall() {
+  const canvasRef = useRef<HTMLCanvasElement>(null)
+
+  useEffect(() => {
+    const reduce =
+      typeof window !== 'undefined' &&
+      window.matchMedia?.('(prefers-reduced-motion: reduce)').matches
+    if (reduce) return // respect users who don't want ambient motion
+
+    const el = canvasRef.current
+    if (!el) return
+    const c2d = el.getContext('2d')
+    if (!c2d) return
+    // Non-null aliases so the render closures keep the narrowed types.
+    const canvas = el
+    const ctx = c2d
+
+    const sprites = PALETTE.map(makeSprite)
+    const dpr = Math.min(window.devicePixelRatio || 1, 2)
+    let W = window.innerWidth
+    let H = window.innerHeight
+
+    // Scale petal count to the viewport so it stays sparse and gentle.
+    const count = Math.round(Math.min(46, Math.max(16, (W * H) / 46000)))
+    const rnd = (a: number, b: number) => a + Math.random() * (b - a)
+
+    const spawn = (initial: boolean): Petal => ({
+      baseX: rnd(0, W),
+      y: initial ? rnd(-H, H) : rnd(-60, -20),
+      size: rnd(16, 38),
+      vy: rnd(22, 52),
+      angle: rnd(0, Math.PI * 2),
+      spin: rnd(-0.9, 0.9),
+      swayAmp: rnd(14, 42),
+      swayFreq: rnd(0.3, 0.85),
+      swayPhase: rnd(0, Math.PI * 2),
+      alpha: rnd(0.5, 0.82),
+      sprite: Math.floor(Math.random() * sprites.length),
+    })
+
+    let petals = Array.from({ length: count }, () => spawn(true))
+
+    function resize() {
+      W = window.innerWidth
+      H = window.innerHeight
+      canvas.width = Math.floor(W * dpr)
+      canvas.height = Math.floor(H * dpr)
+      canvas.style.width = `${W}px`
+      canvas.style.height = `${H}px`
+      ctx.setTransform(dpr, 0, 0, dpr, 0, 0)
+    }
+    resize()
+    window.addEventListener('resize', resize)
+
+    let raf = 0
+    let running = true
+    let last = performance.now()
+
+    function frame(now: number) {
+      if (!running) return
+      const dt = Math.min(0.05, (now - last) / 1000) // clamp big gaps (tab wake)
+      last = now
+      const t = now / 1000
+      ctx.clearRect(0, 0, W, H)
+
+      for (const p of petals) {
+        p.y += p.vy * dt
+        p.angle += p.spin * dt
+        const x = p.baseX + p.swayAmp * Math.sin(t * p.swayFreq + p.swayPhase)
+
+        // Recycle once fully past the bottom.
+        if (p.y - p.size > H) {
+          Object.assign(p, spawn(false))
+          continue
+        }
+
+        ctx.save()
+        ctx.translate(x, p.y)
+        ctx.rotate(p.angle)
+        ctx.globalAlpha = p.alpha
+        ctx.drawImage(sprites[p.sprite], -p.size / 2, -p.size / 2, p.size, p.size)
+        ctx.restore()
+      }
+      raf = requestAnimationFrame(frame)
+    }
+
+    const onVisibility = () => {
+      if (document.hidden) {
+        running = false
+        cancelAnimationFrame(raf)
+      } else if (!running) {
+        running = true
+        last = performance.now()
+        raf = requestAnimationFrame(frame)
+      }
+    }
+    document.addEventListener('visibilitychange', onVisibility)
+    raf = requestAnimationFrame(frame)
+
+    return () => {
+      running = false
+      cancelAnimationFrame(raf)
+      window.removeEventListener('resize', resize)
+      document.removeEventListener('visibilitychange', onVisibility)
+      petals = []
+    }
+  }, [])
+
+  return (
+    <canvas
+      ref={canvasRef}
+      aria-hidden
+      className="petal-fall pointer-events-none fixed inset-0"
+      style={{ zIndex: 20 }}
+    />
+  )
+}

web/src/index.css

@@ -98,6 +98,93 @@ button, a, input {
   padding: 0.1em 0.35em;
   border-radius: 6px;
 }
+/* Top formatting bar: a single line at rest so it stays slim, expanding to
+   reveal every control on hover (or while a control inside it has focus, e.g.
+   the link input or an open popover). The clipped right edge fades out as a hint
+   that more is available. */
+.petal-toolbar {
+  flex-wrap: nowrap;
+  overflow: hidden;
+  max-width: 100%;
+  -webkit-mask-image: linear-gradient(to right, #000 90%, transparent 100%);
+  mask-image: linear-gradient(to right, #000 90%, transparent 100%);
+}
+.petal-toolbar:hover,
+.petal-toolbar:focus-within {
+  flex-wrap: wrap;
+  overflow: visible;
+  -webkit-mask-image: none;
+  mask-image: none;
+}
+
+/* Highlighter mark — inline background swatch behind the text. */
+.petal-prose mark {
+  border-radius: 4px;
+  padding: 0.05em 0.15em;
+  /* color comes from the inline style the Highlight extension writes */
+}
+/* Inserted images: rounded, soft-shadowed, never wider than the column. The
+   selected-node ring matches the rest of the rose theme. Scoped to the editor's
+   own <img> (not a class) since Tiptap doesn't always carry the configured
+   class through to the rendered node. */
+.petal-prose img {
+  max-width: 100%;
+  height: auto;
+  border-radius: var(--radius-card);
+  box-shadow: var(--shadow-soft);
+}
+.petal-prose img.ProseMirror-selectednode {
+  outline: 2px solid var(--color-accent);
+  outline-offset: 2px;
+}
+/* Tables: soft pink grid, rounded outer corners, a tinted header row. Targeted
+   by element within the editor — Tiptap renders the <table> without our
+   configured class, so a `.petal-table` selector would never match. The wrapper
+   Tiptap adds (.tableWrapper) handles horizontal overflow scroll. */
+.petal-prose .tableWrapper {
+  overflow-x: auto;
+}
+.petal-prose table {
+  border-collapse: collapse;
+  width: 100%;
+  margin: 0.4em 0;
+  overflow: hidden;
+  border-radius: var(--radius-input);
+  border: 1px solid var(--color-border);
+  table-layout: fixed;
+}
+.petal-prose td,
+.petal-prose th {
+  border: 1px solid var(--color-border);
+  padding: 0.4em 0.6em;
+  vertical-align: top;
+  position: relative;
+  min-width: 3em;
+}
+.petal-prose th {
+  background: var(--color-surface-alt);
+  font-weight: 700;
+  text-align: left;
+}
+/* The cell being edited / the active selection inside a table. */
+.petal-prose .selectedCell::after {
+  content: '';
+  position: absolute;
+  inset: 0;
+  background: var(--color-surface-alt);
+  opacity: 0.5;
+  pointer-events: none;
+}
+/* The column-resize handle Tiptap draws between columns. */
+.petal-prose .column-resize-handle {
+  position: absolute;
+  right: -2px;
+  top: 0;
+  bottom: -2px;
+  width: 4px;
+  background: var(--color-accent);
+  pointer-events: none;
+}
 /* Placeholder shown on the empty first paragraph (Tiptap Placeholder ext). */
 .petal-prose p.is-editor-empty:first-child::before {
   content: attr(data-placeholder);
@@ -136,6 +223,20 @@ button, a, input {
   animation: petal-suggestion-in 200ms ease both;
 }
 
+/* --- Find & Replace ---------------------------------------------------------
+   In-document search (Ctrl/Cmd+F). Every match gets a soft honey wash; the
+   current match is brighter with a rose ring so it stands out as you step
+   through. Decorations, like every other highlight layer here. */
+.petal-find-match {
+  background: var(--color-highlight, #fff1a8);
+  border-radius: 3px;
+  box-shadow: 0 0 0 1px var(--color-border);
+}
+.petal-find-match-active {
+  background: var(--color-peach);
+  box-shadow: 0 0 0 2px var(--color-accent);
+}
+
 /* --- Spell check ------------------------------------------------------------
    Browser-side nspell flags misspellings with a soft rose wavy underline (a
    gentler take on the classic red squiggle, to fit the pastel palette). Like the
@@ -218,9 +319,17 @@ button, a, input {
    The cozy corner mascot. Gently bobs while awake, settles and sways slowly
    while napping; its speech bubble pops in; little zzz drift up when asleep. */
 .petal-companion {
+  /* Mascot size scales with the viewport width: ~original on a laptop, up to
+     ~2× on a large desktop. Tune the middle (vw) term to taste. */
+  --petal-companion-size: clamp(10rem, 17vw, 20rem);
   animation: petal-bob 3.2s ease-in-out infinite;
   transition: transform 200ms ease;
 }
+/* The Lottie art sits inside the round badge with a little breathing room. */
+.petal-companion-art {
+  width: 88%;
+  height: 88%;
+}
 .petal-companion:hover {
   transform: translateY(-2px) scale(1.04);
 }
@@ -246,10 +355,10 @@ button, a, input {
 }
 
 .petal-zzz {
-  top: 6px;
+  top: calc(var(--petal-companion-size) * 0.04);
-  right: 14px;
+  right: calc(var(--petal-companion-size) * 0.1);
   font-weight: 800;
-  font-size: 1.1rem;
+  font-size: calc(var(--petal-companion-size) * 0.12);
   animation: petal-zzz 2.4s ease-in-out infinite;
 }
 @keyframes petal-zzz {
@@ -382,7 +491,31 @@ button, a, input {
    only the title and the writing itself reach the page. This is Petal's PDF
    path — it uses the browser's own fonts, so CJK renders correctly with no
    server-side font embedding. */
+/* --- Ambient falling petals -------------------------------------------------
+   The WebGL petal layer is a fixed, non-interactive overlay that drifts soft
+   blossoms over the page. Kept gentle (low opacity) so it never competes with
+   the writing. It hides itself in print and is skipped entirely for users who
+   prefer reduced motion (the component renders nothing in that case). */
+.petal-fall {
+  opacity: 1;
+}
+
+/* Honor reduced-motion globally: still the looping ambient animations and drop
+   the petal layer. One-shot entrance/feedback animations are left intact. */
+@media (prefers-reduced-motion: reduce) {
+  .petal-fall {
+    display: none !important;
+  }
+  .petal-bob,
+  .petal-bob-slow,
+  .petal-zzz,
+  .petal-checkpoint-dot {
+    animation: none !important;
+  }
+}
+
 @media print {
+  .petal-fall,
   .petal-no-print,
   .petal-sidebar,
   .petal-suggestion-card,

web/src/vite-env.d.ts

@@ -0,0 +1 @@
+/// <reference types="vite/client" />