Skip to main content
Dari Agent takes an agent project and runs it as a hosted API. Describe the agent in dari.yml, run dari deploy, and you get durable sessions, sandboxed execution, replayable history, webhooks, and versioned releases. A Dari agent can use the managed Pi runtime for a standard coding-agent loop, or a custom runtime server when your own code owns the loop. In both cases, Dari hosts the session lifecycle, records events, manages files, and keeps existing sessions pinned to the version they started on.

​
How It Works

  1. Describe Your Agent in dari.yml: the runtime mode, model configuration, sandbox, tools, skills, and extensions.
  2. Deploy A Version with dari deploy. New versions go behind the same stable agent URL.
  3. Create A Session for each conversation or task. A session is pinned to the active version when it starts.
  4. Send Messages over the CLI or HTTP API. The agent runs in an isolated sandbox and streams events back as work happens.

​
Core Resources

Agents

Stable resources with a version history. Publishing creates new versions.

Sessions

Durable conversations pinned to the agent version active at creation time.

Messages

Per-turn user input and assistant output within a session.

Files And Workspaces

Uploaded files and sandbox workspace outputs associated with sessions.

​
Start Here

Agent Quickstart

Publish your first agent and send a message.

Manifest Reference

Learn the full dari.yml contract.

Runtime Modes

Choose between the managed Pi runtime and a custom runtime server.