Recursive Language Models (GitHub Summary)

Recursive Language Models — GitHub Repository Summary

Repository: https://github.com/alexzhang13/rlm

Stars: 3,569 | Forks: 650 | Issues: 77

License: MIT

Language: Python

Maintainer: MIT OASYS lab

Overview

The official implementation of Recursive Language Models (RLMs) provides an extensible inference engine that wraps standard API-based and local LLMs with recursive REPL-based context handling. Installable via pip install rlms [Source: sources/rlm-github.md].

Key Features

Drop-in API

from rlm import RLM
 
rlm = RLM(
    backend="openai",
    backend_kwargs={"model_name": "gpt-5-nano"},
    verbose=True,
)
print(rlm.completion("Print me the first 100 powers of two, each on a newline.").response)

Multiple REPL Environments

Environment	Type	Isolation	Notes
`local` (default)	Non-isolated	Host process	Uses Python `exec`; shares venv; safe for benchmarking
`docker`	Semi-isolated	Container	Default `python:3.11-slim` image
`modal`	Isolated	Cloud	Modal Sandboxes
`prime`	Isolated	Cloud	Prime Intellect Sandboxes (beta)
`daytona`	Isolated	Cloud	Daytona Sandboxes
`e2b`	Isolated	Cloud	E2B Sandboxes [Source: sources/rlm-github.md]

Model Provider Support

OpenAI, Anthropic
Router platforms: OpenRouter, Portkey
Local models via vLLM (OpenAI-compatible client) [Source: sources/rlm-github.md]

Logging & Visualization

In-memory metadata: completion.metadata holds full trajectory (config + iterations + sub-calls)
Disk logging: RLMLogger(log_dir="./logs") writes JSONL for the visualizer
Visualizer: Node.js/shadcn/ui app to inspect code, sub-LM, and root-LM call trajectories [Source: sources/rlm-github.md]

Minimal implementation: https://github.com/alexzhang13/rlm-minimal — stripped-down reference for building on top of RLMs [Source: sources/rlm-github.md]

Citation

@misc{zhang2026recursivelanguagemodels,
      title={Recursive Language Models},
      author={Alex L. Zhang and Tim Kraska and Omar Khattab},
      year={2026},
      eprint={2512.24601},
      archivePrefix={arXiv},
      primaryClass={cs.AI},
      url={https://arxiv.org/abs/2512.24601},
}

[[recursive-language-models]] — Main concept page
[[rlm-paper]] — arXiv paper summary
[[alex-l-zhang]] — Lead author / maintainer