Production-quality demo of an event-driven AI orchestrator between a live transcription stream and an application's REST API. It is intentionally not a chatbot: transcript chunks become validated pending actions, and human approval generates exact API calls without sending them.

docker compose up --build

Open http://localhost:5173 for the dashboard and http://localhost:9000/docs for the API.

The planner runs in hybrid mode. If ANTHROPIC_API_KEY is present, it uses Claude with structured tool output. Without a key, it uses deterministic intent extraction so the demo and tests remain offline-friendly.

Approved actions generate calls for the real Wed backend API without executing them. Configure:

export WED_API_BASE_URL="http://host.docker.internal:8000"
export WED_DEFAULT_PARTNER_ID="partner_uuid_for_guest_reservation_requests"
export SONIOX_API_KEY="your_soniox_api_key"

Use Start Mic in the dashboard to record microphone audio. When you stop recording, the backend transcribes it with Soniox and feeds the completed transcript segments into the AI orchestrator.

Languages

Python 80.4%

TypeScript 14.3%

CSS 4.7%

HTML 0.3%

Dockerfile 0.3%