Skip to main content
Fleet RLM is a terminal-based assistant for work that needs reasoning, Python, and files. You ask a question, watch the work stream into your terminal, and return to the saved Session later. Fleet RLM combines three pieces:
  • DSPy dspy.RLM runs the recursive language model (RLM) reasoning loop.
  • Daytona Sandboxes run every line of model-authored Python, outside the Fleet process.
  • A FastAPI backend streams progress over Server-Sent Events (SSE) and persists Sessions, Turns, and Artifacts.
The current release is fleet-rlm 0.7.11. It requires Python 3.11 to 3.13 and is certified on exactly DSPy 3.4.0.

​
What you can do

  • See the work as it happens. The terminal shows reasoning, generated Python, tool activity, and results in one timeline.
  • Continue a Session. Committed conversation history survives a restart when a database is configured.
  • Work with files. Attach local files to a Turn, browse workspace files, and download committed Artifacts.
  • Delegate bounded sub-investigations. Fleet enables DSPy semantic calls and isolated child RLMs for independent work. Set rlm.recursion_enabled = false for native-only operation.
  • Drive it from your own client. The same backend runs without the terminal and exposes a small HTTP and SSE API.

​
How the pieces fit

The backend prepares each request, runs model-authored Python in Daytona, and streams progress to the terminal. Fleet commits the answer and any Artifacts only after the Run settles.

​
Next steps

Quickstart

Install Fleet RLM and run your first live Session.

How Fleet RLM works

Follow a Turn from request to committed result.

Use the terminal

Slash commands for files, Sessions, Skills, and settings.

HTTP API

Sessions, Turns, files, and the SSE stream contract.
Last modified on October 9, 2026