pi coding agent

pi is a terminal coding agent, and orboto treats it as a first-class teammate. There are three integration paths - they compose, and all three respect the same permissions and agent rules as every other surface.

1. Interactive sessions via MCP

For a pi session you drive yourself, connect orboto's MCP server - pi speaks MCP natively through its pi-mcp-adapter package (pi install npm:pi-mcp-adapter). Then register orboto in ~/.pi/agent/mcp.json:

{
  "mcpServers": {
    "orboto": {
      "url": "https://your-orboto-host.example.com/mcp",
      "auth": "oauth",
      "oauth": {
        "scope": "mcp api offline_access",
        "clientName": "Pi"
      }
    }
  }
}

No API key to paste: on first use, pi completes an OAuth login in your browser against your orboto (dynamic client registration - nothing to configure server-side). The session then has the curated orboto toolset: tickets, search, docs, time tracking, and the orboto_api_call escape hatch to every other endpoint. See MCP server setup for toolsets and details.

2. The orboto CLI inside any session

Autonomous work runs best over the orboto CLI - a single static binary the agent calls like any other tool: claim a ticket, log time, query with OQL, finish work. It authenticates with a bot API key and is what the runner's bootstrap prompt points the agent at. The CLI daily workflow page lists the whole loop.

3. Full autonomy with pi-runner

orboto pi-runner (built into the CLI) supervises a headless pi session that pulls tickets on its own - the one-command setup is Your first AI worker, the always-on version is Run an agent fleet.

Model providers

pi brings its own providers (Anthropic, OpenAI, and more). For local or self-hosted models, the pi-lm-providers extension adds Ollama (local + Ollama Cloud) and LM Studio with live model discovery - the runner setup wizard installs it for you when you pick one of those.