Summary

This repository offers a minimal Recursive Language Model (RLM) scaffold using Claude Code. It tackles the challenge of processing inputs exceeding LLM context windows by enabling programmatic decomposition and recursive self-calling over document chunks. This benefits Claude Code users requiring analysis of extensive text data.

How It Works

The architecture mirrors the RLM paper: a root LLM (Claude Opus 4.5) orchestrates sub-LLM calls (Claude Haiku) via a persistent Python REPL. This REPL functions as the external environment, managing state and providing utilities for chunking and searching. This approach allows recursive processing of inputs significantly larger than standard context limits.

Quick Start & Requirements

• Installation: Clone the repository (git clone https://github.com/Brainqub3/claude_code_RLM.git) and cd into it.
• Prerequisites: Claude Code account (Anthropic CLI), Python 3.
• Usage: Initiate Claude Code within the repo directory (claude), then execute the RLM skill (/rlm). Follow prompts for context file path and query.
• Organization: Large context files should reside in the context/ directory.
• Links: RLM Paper

Highlighted Details

• Handles inputs up to two orders of magnitude beyond model context windows.
• Employs a persistent Python REPL for stateful operations and utility functions.
• Leverages Claude Opus 4.5 as the orchestrator and Claude Haiku for efficient chunk analysis.

Maintenance & Community

No specific details regarding contributors, sponsorships, community channels, or roadmaps are present in the README.

Licensing & Compatibility

The README directs users to a LICENSE file for details. The specific license type and implications for commercial use or closed-source integration are not explicitly stated.

Limitations & Caveats

This is a minimal implementation, not production-ready. Use of Claude Code's --dangerously-skip-permissions mode necessitates careful setup, isolated environments, and awareness of risks due to unconfirmed command execution.