Live Recursive Agents
RLM Code v0.1.13 expands the experimental native coding agent with live recursive agents.
The root agent can now start multiple child agents concurrently, communicate with them, collect their
findings, apply the solution, and run verification—all under the configured sandbox and approval policy.
What’s new
- Concurrent child agents from the persistent Python workspace
- Parent and child messaging and steering
- Wait, cancel, delete, and follow-up controls
- Live agent hierarchy and activity in the terminal
- Complete root-and-child trajectory recording
- Replay with rlm-code agent replay
- Session resume with child records, results, supported Python state, and undelivered messages
- Shared sandbox, approval, audit, and budget controls across the hierarchy
Install
pip install --upgrade rlm-code==0.1.13
Experimental status
The native agent runtime remains experimental. Active child agents are settled when the RLM Code process
exits. Background continuation, daemon supervision, and automatic crash recovery are planned for a future
release.
Docker or Apple Container is recommended for isolation. The local runtime is intended for development in
trusted environments.