Turn written text into spoken audio and keep a permanent, browsable library of everything you've generated. EchoRecall is a small web app: build a list of text, pick a voice and format, and generate a whole batch at once — every successful clip is saved so you can replay, retag, download, or delete it later without regenerating.

• Generation workspace — a resizable two-pane dashboard (queue list + metadata editor): build a queue by typing a line or uploading a .txt, then search, filter, multi-select, and choose which columns to show. A header toolbar drives upload, previous/next, generate, and save/open of the queue as a local .echoqueue.json file. Generate targets the checked rows (else the whole queue), isolates per-item failures, removes the successes from the queue, and offers that run as one .zip.
• Voices, models & formats — choose a voice and model, and render to MP3, WAV, FLAC, Opus, AAC, or PCM.
• Standards-based metadata — attach title, artist, album, genre, comment, languages, a recording date, and repeatable custom text/URL tags, written as ID3v2.4 / Vorbis comments where the format supports them.
• Persistent library — every generation is stored (SQLite + on-disk audio under data/) and survives restarts; replay makes no new provider call. Files are saved with human-readable, title-slugged names under a YYYY/MM/DD folder.
• Library workspace — the same two-pane layout (results table + an audio-tags panel with previous/next), plus a waveform review player with zoom and loop regions (playback aid only — it never modifies the audio). Server-side search, sort, filter (voice/format/date), and pagination; rename, retag, delete, or bulk-clean by filter.
• Personalize — light/dark theme and language (English / Hungarian, Hungarian default), persisted per browser.
• In-app OpenAI key — optionally set/replace/clear the OpenAI key from Settings (a modal opened from the toolbar or header); it is encrypted at rest and never returned to the client.
• Versioned — the running app version is shown in the header and bumped with one release command.

There is no built-in authentication — secure network access yourself (localhost, reverse proxy, or VPN).

TypeScript (strict) on Node.js · Nuxt 4 (Vue 3 + Nitro) · @nuxt/ui v4 (+ color-mode, dashboard primitives) · @nuxtjs/i18n (en/hu) · wavesurfer.js (Library waveform review, display/playback only) · OpenAI TTS · multi-format audio with ID3/Vorbis tagging via taglib-wasm (pure WASM — no system binary) · SQLite (better-sqlite3) + filesystem audio under data/ · Vitest + @nuxt/test-utils · Docker Compose. A framework-agnostic core lives in src/core/; the Nuxt server (server/) and UI (app/) are adapters over it.

• Node.js 22.22.2 — pinned in .mise.toml / .node-version. Using mise (mise install) is recommended; the native better-sqlite3 addon must be built against this version. (Node ≥ 22.6 is required for the WASM tagger.)
• pnpm 10 (corepack enable will provide it).
• An OpenAI API key with TTS access — set it in .env or in Settings (the in-app modal).

pnpm install
cp .env.example .env          # then set NUXT_OPENAI_API_KEY=sk-... (or set it in Settings)
pnpm dev                      # http://localhost:3102

Environment variables (server-only; never commit real keys):

Default tag values (Artist, Album, Genre, Comment, Languages — Title is never defaulted) are configured in Settings ("Default tag values") and persisted in the app; they are no longer set via environment variables. Upgrading from a release that used NUXT_DEFAULT_TAG_*? Those values are not migrated — re-enter them once in Settings.

pnpm test            # core unit + server integration (TTS provider mocked — no live calls)
pnpm test:component  # Vue component tests (@nuxt/test-utils)
pnpm test:adapters   # GATED: real taglib-wasm tagger (+ optional live OpenAI if keyed)
pnpm typecheck       # nuxt typecheck
pnpm lint            # eslint

The default suites mock the TTS provider and tagger at the core port boundary, so they never touch the network or a system binary. The real WASM tagger is exercised only by the opt-in test:adapters suite.

pnpm bumpp <patch|minor|major>   # omit the type to choose interactively

bumpp bumps the single version in package.json and creates a release commit + tag. nuxt.config.ts reads that version into runtimeConfig.public.appVersion, which the header displays — so the shown version updates with no further edit.

# Reads NUXT_* vars from .env; serves on http://localhost:3102
docker compose up --build -d

docker-compose.yml passes the OpenAI key and NUXT_APP_SECRET through from .env (both optional), and bind-mounts ./data so the SQLite database and generated audio persist across container recreation. No extra system packages are needed — better-sqlite3 is built in the image and the tagger ships as WASM via npm. Stop with docker compose down.

src/core/      framework-agnostic domain: tts/, library/, settings/, shared/ (ports + use-cases)
server/        Nitro API routes (server/api/, incl. settings/) + DI container & error mapping
app/           Nuxt UI: components/ (generate, library, settings, common), composables/, pages/
i18n/locales/  en.json · hu.json
tests/         unit/ · integration/ · component/ · adapters/ (gated)
data/          runtime SQLite DB + audio/ (gitignored; Docker volume)
specs/         feature specs, plans, data models, API contracts, quickstarts

The latest feature — the dashboard redesign (a shared resizable two-pane workspace for Generate and Library, queue search/filters/columns/multi-select, local save/open of the queue, the Library waveform review player, and Settings in a modal) — is specified in specs/005-dashboard-redesign/ (see its quickstart.md for end-to-end validation steps). The batch studio, formats, tagging, library, settings, defaults, and versioning are specified in specs/002-studio-enhancements/; the original single-generation baseline lives in specs/001-tts-generation-library/.