Skip to content

README.md

Latest commit

 

History

History
49 lines (34 loc) · 1.66 KB

README.md

File metadata and controls

49 lines (34 loc) · 1.66 KB
Raw

MOSS Voice Agent — Vercel AI Gateway

Realtime voice agent using Vercel AI Gateway with MOSS as the knowledge base. Speak a question — the agent searches your MOSS index and answers out loud.

Architecture

Browser (useRealtime) ──WebSocket── Vercel AI Gateway ── gpt-realtime-2
        │                                   │
        │   tool call: search_knowledge_base │
        └─── POST /api/token ───────────────►│
                                         MOSS index (local)
  • POST /api/token (empty body) — mints a short-lived WebSocket token via the gateway
  • POST /api/token ({ query }) — executes MOSS search; uses local in-memory index loaded at startup

Security note: /api/token is unauthenticated only when ALLOW_UNAUTHENTICATED_DEMO=true is set (demo). Before deploying publicly, add a session/cookie check so arbitrary callers cannot mint Gateway tokens or query your index.

Setup

1. Install dependencies

Requires Node.js ≥ 22 (ai@7 and @ai-sdk/gateway@4 require it).

npm install

2. Add credentials

cp .env.example .env
Variable Where to get it
MOSS_PROJECT_ID moss.dev dashboard
MOSS_PROJECT_KEY moss.dev dashboard
MOSS_INDEX_NAME Name of the index to search
AI_GATEWAY_API_KEY Vercel AI Gateway → API Keys

3. Run

npm run dev

Open http://localhost:3000 and tap the orb to start talking.