Repository navigation
First public release. The repository starts a fresh history at this version.
Changed
- Vizier is now only a display. The server no longer listens, transcribes or calls a model: it shows what is
posted to its API. Removed: the/ears/*audio endpoints, the built-in speech-to-text and the Claude bullet writer,
the take review,scripts/desk-ears.shand every credential setting (ANTHROPIC_API_KEY,VIZIER_CREDS,
VIZIER_MODEL,VIZIER_USER_NAME,VIZIER_RULES_EXTRA,VIZIER_KEYTERMS). - Take recording is off unless
VIZIER_CAMis set (it was on, against a default stream URL). - The app's CLIENT tab is gone, and so is its built-in list of author names to hide: list them in the app's config
as"hideLabels": [...].
Added
- Display API:
POST /lines,GET /transcript,POST /bullets,POST /status,POST /script,
POST /prompter/note, and an OpenAPI 3.1 spec (docs/openapi.yaml, served at/openapi.yaml). VIZIER_TOKEN: a bearer token required on every POST.connectors/mcp/: an MCP server (standard library only) with tools to post lines and bullets, set the script,
status and note, and read the transcript;VIZIER_MCP_PUSH=1pushes new lines into a Claude Code session.connectors/livekit/: a bridge from a LiveKit room's standard transcription to/lines, and a page that publishes
both sides of a call (mic =me, screen-share audio =them) into a room.examples/(curl, a minimal agent loop) anddocs/architecture.md.- A window that connects mid-call receives the call's last 8 lines; the app and the browser view dedupe by
seq. - Desktop app: talk to an agent in a LiveKit room. Alt+E toggles Ears (your mic into the room), Alt+M toggles
Mouth (the agent's replies); both persist across restarts, publish as participant attributes, and every switch is
logged throughPOST /talk/log. The agent's audio is muted while you speak (barge-in). - Prompter: global keys (Alt+↓/↑ from any app; plain Tab, ↑, ↓, ←, → global, released and retaken with Alt+K),
Alt+T next tab, a key legend, cue styling for[show]lines, 32 px type. - Prompter driving API for agents:
POST/GET /prompter/position(where the reader is),POST /prompter/goto(jump to a
part and line),POST /prompter/follow(Alt+F toggles speech-follow),POST /prompter/snap(the app writes a
picture of the prompter, since the window is hidden from screen capture).
SHA-256: 188f14a560288376fd241c18c445415a5db915ce1c0c990fea32826f82b495b4
Signed with the certificate in signing/.