docs

fleet-rlm Documentation

fleet-rlm provides secure, cloud-sandboxed recursive language model workflows with a Web UI, API, and optional MCP server. The default product path is Modal-backed. The Daytona path is experimental, but it plugs into the same workspace and transport contract instead of acting like a separate app.

This documentation is for both:

• users operating fleet-rlm locally or in deployment workflows
• contributors building integrations, extending runtime behavior, or maintaining the codebase

Quickstart

uv init
uv add fleet-rlm
uv run fleet web

Then open http://localhost:8000.

Next steps:

• Installation details
• Runtime settings
• Deploying the API server

Choose Your Path

Use the product

• Installation
• Runtime settings
• LiteLLM proxy model availability
• Deploying the API server
• Troubleshooting

Contribute to the project

• Contributing guide

Build integrations

• HTTP and WebSocket API
• Python API
• CLI reference
• Using the MCP server

Understand architecture

• Architecture overview
• Concepts
• User interaction flows
• Component UML

Documentation Map (Diataxis)

• Tutorials
• How-to Guides
• Reference
• Explanation

Source-of-Truth Policy

When docs conflict with implementation, treat these as authoritative:

• CLI truth: uv run fleet-rlm --help and uv run fleet --help
• API truth: openapi.yaml
• WebSocket truth: src/fleet_rlm/api/routers/ws/api.py

Archive Note

Historical docs are archived and non-operational:

• Legacy planning docs are stored under the local-only plans/archive/docs-legacy/ path.