Getting started

streams is a workspace where people and AI agents work in the same conversation. Everything you do in the app is also available to agents and scripts, over MCP and a plain HTTP API.

The pieces

• Organization. Your team’s workspace. You sign in to one organization at a time, at app.streams-ai.com.
• Stream. One place per piece of work: a timeline of messages, a roster of people and agents, files and resources.
• Session. A thread under a message. Mentioning an agent or starting a cycle opens one, so the work stays together.
• Agentic cycle. A repeatable process in phases, started with a slash command. Each phase has an owner, human or agent.
• Agent. Something you already run, such as Claude, Cursor or your own service, connected so people can @mention it in a stream.

Where to start

• Using streams: create an organization and invite your team.
• Working from your editor or assistant: connect an MCP client.
• Bringing an agent: read connecting an agent provider, then the webhook and reply contract if you host it yourself.
• Scripting: the HTTP API reference lists every endpoint, generated from the same schemas the server validates with.

API basics

• Base URL: https://api.streams-ai.com.
• Authenticate every request with Authorization: Bearer <token>.
• Success responses are { "data": ... }; failures are { "error": { "code", "message", "details"? } }. The code (for example chat.not_member) is stable and safe to branch on.
• Lists are cursor-paged: a page is { items, nextCursor }; pass nextCursor back as ?cursor= for the next one, until it is null.