@neural_avb: Shipping the latest fast-rlm fast-rlm lets your LLMs work inside a RLM harness, exploring massive contexts inside a REP…
Summary
fast-rlm is an open-source Python tool that allows LLMs to operate within an RLM harness for recursive subagent calls, with features like spend limits and live log streaming.
View Cached Full Text
Cached at: 07/02/26, 06:25 PM
Shipping the latest fast-rlm
fast-rlm lets your LLMs work inside a RLM harness, exploring massive contexts inside a REPL + sandbox + recursively call subagents. Supports tools and subagent level structured IO.
What’s new:
-
Spend limits that actually stop runs. You can simply state a soft $ budget for a RLM call, and it will respect it.
-
Stream logs step by step as it happens. Great for your coding agent to monitor an RLM run and report results “live”
-
More patches like better logging, event listeners, explicit KV Cache hits for Anthropic models, and better starting defaults.
Fast RLM is pip installable, and repo is open and MIT.
Similar Articles
@neural_avb: RLMs can now access MCP servers with `fast-rlm` - Connect any MCP via stdio or http - RLM accesses all MCP tools, resou…
fast-rlm enables reinforcement learning models to access MCP servers via stdio or HTTP, allowing tool use and resource fetching with results saved as Python variables in the REPL to save input tokens.
@neural_avb: One of the the coolest RLM trajectories that made me go "woah" RLMs (Minimax M3) launching subagent swarms with clear p…
Neural_avb highlights how Minimax M3's RLMs use subagent swarms with pydantic contracts for type checking and schema validation, reducing hallucination rates and failed subagent calls.
@TDataScience: Follow along @neural_avb's all-in-one deep dive to learn "what recursive language models (RLMs) are, why they are winni…
An educational deep dive into recursive language models (RLMs), explaining what they are, why they are winning long-context benchmarks, and how they differ from existing agentic harness designs like ReAct or CodeAct, using a simple case study.
@neural_avb: Locally generating GRPO-like rollouts with my SLM, and using this tiny RM as the rubric. Next I'll be RL training on fr…
Neural_avb releases a lightweight Answer-eq Reward Model for RL training on QA tasks, claiming 80% agreement with external judge LM and faster than F1/ROUGE/BertScore.
@samhogan: RLMs pretty much solved context btw You can shove tens of millions of tokens into a good RLM harness and it just works.…
A developer shares their experience with Recurrent Language Models (RLMs), claiming they effectively handle extremely long context windows with tens of millions of tokens, representing a significant advancement in context handling capabilities.