OMA × Next.js Starter

Deploy a multi-agent app in one click. A researcher agent gathers information and a writer agent composes an article, orchestrated by open-multi-agent (runTeam()) and streamed to a chat UI via the Vercel AI SDK.

What this shows

The Vercel AI SDK gives you a single agent loop and a streaming UI. open-multi-agent adds the layer the AI SDK deliberately leaves out: multi-agent orchestration. A coordinator decomposes your goal into a task DAG, agents run in dependency order with shared memory, and the result streams back through the AI SDK.

User topic
  │
  ▼
app/api/chat/route.ts
  ├─ open-multi-agent runTeam()  →  coordinator → researcher → writer (shared memory)
  └─ Vercel AI SDK streamText()  →  streams the article to the browser
  ▼
Chat UI (app/page.tsx, useChat)

Deploy

Click the button above, set GEMINI_API_KEY, and deploy. Get a free key at Google AI Studio.

The agent team runs inside a serverless function. maxDuration defaults to 60s to fit Vercel's Hobby (free) tier. Heavier topics may need Vercel Pro (up to 300s).

Run locally

npm install
cp .env.example .env.local   # then add your GEMINI_API_KEY
npm run dev

Open http://localhost:3000, enter a topic, and watch the team work.

Use a different model

The demo runs on Google Gemini because it is fast, has a free tier, and is reachable from every Vercel region. Both open-multi-agent and the AI SDK speak the OpenAI-compatible API, so switching to OpenAI, Anthropic, DeepSeek, Groq, or a local model is a few-line change to the constants in app/api/chat/route.ts. See the open-multi-agent provider docs.

Key files

File	Role
app/api/chat/route.ts	open-multi-agent orchestration + AI SDK streaming
app/page.tsx	Chat UI (useChat)

License

MIT