web

⚙️ voice-assistant config panel

The web admin for voice-assistant — a Nuxt 4 SSR app (Nuxt UI) that runs its own Nitro server and edits the same SQLite database the assistant process uses. This is the way the assistant is configured: no .env edits, no redeploys for config.

What it edits

Page	Table(s)	What
Integrations	integrations	OpenAI (mandatory), Home Assistant, Telegram — install / enable / connection-test. Secrets (keys, tokens) are stored here, masked in the UI.
Tools	settings	Built-in tool gates (memory, reminders) + weather (units, location).
Prompts	prompts	Every system / tool prompt, editable, with restore-to-default.
Users	users + identities	Principals and their devices: Telegram chats, HTTP tokens, voice device tokens.
HA Voice / HTTP API	settings	Realtime enable switch (+ pacing / idle) and the per-endpoint HTTP toggles.

When changes take effect:

• Config (integrations, settings, prompts) is read at the assistant's startup — these apply on the next restart, and the UI says so on save.
• Users & devices (users, identities) are read per request, so they apply live — add a user or device and it can authenticate immediately.

Secrets

Secrets now live in the DB, entered through this panel, not in .env: the OpenAI API key, the Home Assistant token, and the Telegram bot token are stored in the integrations table and masked in every GET (the UI shows "leave blank to keep"). The assistant's .env carries only process/infra knobs (DB path, TZ, ports). Voice/HTTP bearer tokens are never stored raw at all — only their sha256 hash, as identities rows.

Schema ownership

voice-assistant owns the schema and migrations. This app never migrates — if a table doesn't exist yet, start the assistant once against the DB so it creates them (a fresh-DB run also seeds the editable prompts). Reads tolerate missing tables; writes return 503 until they exist.

server/utils/settable.ts mirrors the assistant's src/settings/settable.ts. Keep the two in sync until they're extracted into a shared module.

Dev

npm install
VA_DB_PATH=../data/assistant.db npm run dev   # this is the default

Open http://localhost:3000. The dev server talks to the sibling repo's data/assistant.db.

Build / run

npm run build
PORT=3000 VA_DB_PATH=/data/assistant.db node .output/server/index.mjs

Deploy

Built as its own image and wired into the home-infra stack (port, Caddy + tinyauth route, shared data/ volume, depends_on: voice-assistant). Authentication is handled by the reverse proxy — this app has no built-in auth, so never expose it without one in front.