Source: vercel.com/changelog/humans-and-ampersand-now-available-on-ai-gateway
Ampersand is built to take part in group conversations such as team chats. On each turn it chooses between speaking and staying quiet, aiming to understand individual and shared goals and to help resolve disagreements. A contribution can be a clarifying question, relevant context, or a decision to wait for people to reply.
The application defines two caller-supplied function tools: send_message, which posts a contribution, and pass, which keeps the conversation quiet until the next human message. The application runs both tools and records their results. Tool choice should stay at its default, because Ampersand accepts auto and none but rejects required and named choices. The model is called as humansand/ampersand-v0.1 through the AI SDK, the OpenAI-compatible Chat Completions API, or a coding agent.
Vercel's example saves responses, history and delivery receipts in durable storage for deployed chat apps. Uncertain deliveries are retried with the saved action and the same receipt, messages are processed in order per conversation, and conversation access is checked before posting. To use the model in Claude Code, Codex or Cursor, run npx vercel ai-gateway setup with the latest Vercel CLI.
Why it matters if you build agents
If you build chat agents, Ampersand can choose not to respond, so your code should treat a pass action as a normal outcome rather than an error. Keep tool choice at auto or none, since required and named choices are rejected. AI Gateway reports that it adds no markup and no platform fee on inference, including Bring Your Own Key requests.
Get stories like this every weekday
One short email with the day's agent tooling news.