Arc Forge Labs Docs v2026.10.2
Dictate

Frontend ↔ engine wiring status

What the shipped UI (ui/src) talks to, and what still needs backend work to be fully real. The UI runs fully mocked with no backend (see ui/README.md — the fast loop); this file tracks the gaps between that mock and the live engine.

Bridge: ui/src/ipc.js → ui_server.py HTTP (loopback, token-auth) for state, plus Tauri commands in ui-shell/src-tauri/src/lib.rs for window/OS actions.

Wired and working

UI capability Backend
Hydrate + live updates GET /api/state, SSE GET /api/events
Model / shortcut / theme / startup / device PATCH /api/config
Hotwords add/remove POST / DELETE /api/hotwords
History list / clear GET / DELETE /api/history
Note recording (start/stop/pause/resume/toggle) POST /api/notes/*
Provider API keys POST / DELETE /api/api-keys for advanced/BYO-key paths
Doctor POST /api/doctor
Export note as Markdown Tauri save_text_file (OS dialog) ← ipc.saveTextFile; browser-download fallback
Privacy pill (on-device ↔ online) PATCH /api/config model swap for advanced/BYO-key paths; Dictate Pro should use entitlement-backed hosted access
Update check + run + restart GET /api/update-status, POST /api/update, Tauri restart_app ← ipc.restartApp

Needs wiring (gaps against the new UI)

  1. Update background-prepare. The Update pill has a available → preparing → ready progression, but live mode has no prepare step — clicking runs download+pkexec+install in one go via POST /api/update. To match the staged UX, add POST /api/update/prepare (download the .deb in the background, report progress), have /api/update install the prepared file, then the UI calls restart_app. Optionally push download progress over the SSE /api/events channel so the pill can show a real bar.

    • Frontend marker: App.jsx useEffect launch flow has a TODO(backend): /api/update/prepare note; mock simulates the staging.

  2. Skip persistence. App.jsx stores skippedVersion in localStorage (per-webview, lost on cache clear). Move it to a server-side pref (ui-prefs.json via a config route) so "Skip this version" is authoritative and survives reinstalls. "Later" (dismiss) is correctly session-only.

  3. Update path E2E. The download → pkexec apt-get install → restart_app flow is unit-tested with mocks only (tests/test_update_status.py). First real exercise needs a newer published release: verify the polkit prompt, the atomic file swap, and that the new shell + engine come up together.

  4. Local Meeting model lanes. The capture home exposes a plain Meeting action and routes it to strict backend meeting endpoints. Live meeting mode now fails closed unless the selected backend supports speaker-attributed output. The remaining gap is implementing and benchmarking the local DiariZen, Sortformer, and pyannote speaker-attribution lanes from docs/TRANSCRIPTION_PLAN.md.

  5. WhisperX backend. UI metadata exists for whisperx, but docs/TRANSCRIPTION_PLAN.md says not to make WhisperX the main meeting stack. Keep it advanced/experimental unless that plan changes.

  6. Dictate Pro / subscriptions. Architecture exists in docs (dictate-pro-subscription-architecture.md, msstore-in-app-subscriptions.md, archived provider cost notes) but there is no frontend surface yet. Production P0: public Upgrade to Pro on https://arcforge.au/download/dictate is live but still links to /login until Stripe product + account entitlements are wired (see docs/archive/goals-2026-07-05.md for the historical goal note and arc-forge-website/STATUS.md for website state). To wire in this repo: entitlement/subscription state in GET /api/state, a Pro affordance in the UI, checkout/portal deep links from the app, MS Store IAP (Windows) where applicable, and service-side entitlement checks before hosted meeting upload. Do not expose Arc Forge's hosted provider key or a primary switchable-backend workflow in the Pro UI.

Edit this page on GitHubMarkdownSource fe527b9