Recursive Language Model

RLM (Recursive Language Model)

Process arbitrarily long documents by treating them as an external environment: keep the document in a persistent Python REPL, search and chunk it there, and recursively delegate chunk-level analysis to subordinate agents. Never paste large content into the main conversation.

All paths below are relative to this skill's directory. Requires Python 3.8+ and an agentic harness with code execution and subordinate agents.

Workflow

1. Load the document into the REPL: python3 scripts/rlm_repl.py init <path/to/context.txt> (verify with ... status)
2. Scout the content: python3 scripts/rlm_repl.py exec -c 'print(peek(0, 3000))' and locate relevant sections with grep(pattern) before reading anything in bulk.
3. Chunk if needed: python3 scripts/rlm_repl.py exec -c 'print("\n".join(write_chunks("./rlm_chunks", size=200000)))'
4. Delegate each relevant chunk to a subordinate agent using the prompt profile in rlm-subcall.md, passing the query and the chunk file path. Prefer JSON-structured replies.
5. Synthesize the sub-agent results in the main conversation and answer.
6. Clean up: python3 scripts/rlm_repl.py reset

Reference

• README.md — full REPL command and helper-function reference, architecture, compatible harnesses.
• EXAMPLE.md — complete worked example on a large document.
• rlm-subcall.md — subordinate agent prompt profile.
• scripts/rlm_repl.py — the persistent REPL (state stored locally via pickle; context size limited by memory).