RLM Agents
https://github.com/hmbown/rlmagents- Category
- AI Agents
- Rank
- No. 1571Tools index
Previous survey · No. 1578 ·
- Pricing
- Open Source
- Type
- TOOL
- Builder
- Hmbown
- GitHub
- 49 stars
- Latest release
- rlmagents==0.0.5
- Date
About
RLM agent harness built on top of Deep Agents for running reinforcement-learning-style multi-agent workflows.
What it does
RLM Agents tackles large analysis tasks by keeping data in session state, running Python during an iterative reasoning loop, and making recursive model queries. It also tracks evidence, isolates working contexts, and supports reusable analysis recipes.
Why it's ranked here
The strongest case is its unusually complete investigation workflow. A Python API, interactive terminal interface, resumable conversations, evidence tracking, recursive queries, and protocol integration make it useful beyond a research prototype. The substantial dependency and execution surface demands careful setup.
What's good
Separate models can handle primary reasoning and recursive queries. Isolated contexts prevent unrelated work from sharing session state. Large tool results can load into those contexts automatically. Recipe validation, citations, planning, memory, filesystem tools, sub-agents, and configurable tool profiles support repeatable investigations.
Tradeoffs
The harness depends on LangChain and LangGraph, so adopters inherit a substantial framework stack. Its core loop runs generated Python, while broader agent tooling can execute shell commands and modify files. Automatic approval removes human checks and can permit arbitrary commands. Remote execution requires separate sandbox configuration.
How to use it well
Use it for large repositories, long documents, and multi-step research where evidence must survive several tool calls. Start interactively, preserve useful threads, then turn stable procedures into recipes or scripted Python integrations. It suits engineers comfortable governing code execution. It does not supply the underlying language model or sandbox service.
Technical notes+
libs/rlmagents/rlmagents/__init__.py exports create_rlm_agent, RLMMiddleware, and RLMSessionManager. libs/cli/deepagents_cli/main.py defines interactive and non-interactive CLI modes, thread management, model selection, shell allow lists, auto-approval, and remote sandbox choices; libs/rlmagents/deepagents_cli/main.py contains a bundled counterpart. libs/acp/deepagents_acp/server.py implements an Agent Context Protocol server around compiled LangGraph agents, including sessions, modes, cancellation, plan updates, and streamed tool-call reporting. Makefile manages lock checks across the monorepo and specifies Python 3.14 for libs/acp, versus Python 3.12 for the other listed packages. README.md documents the public RLM loop, package layout, Python API, CLI, recipe tools, evidence tracking, and multi-context isolation.
Observed
- License
- MIT
- Primary language
- Python
- Install surface
- Published package installable with pip install rlmagents
- Runtime requirement
- Python 3.11 or newer is stated in the project documentation
- Interfaces
- Python library, interactive terminal UI, one-shot CLI, and Agent Context Protocol integration
- Core frameworks
- Built with LangChain and LangGraph
- Repository structure
- Monorepo containing the main library, standalone CLI, ACP integration, evaluation tooling, and an upstream-compatible Deep Agents SDK
Read from README.md, Makefile, .github/scripts/check_extras_sync.py, libs/cli/deepagents_cli/app.py, libs/cli/deepagents_cli/main.py, libs/acp/deepagents_acp/server.py, libs/acp/deepagents_acp/__main__.py, libs/cli/deepagents_cli/__init__.py, libs/cli/deepagents_cli/__main__.py, libs/rlmagents/deepagents_cli/app.py, libs/rlmagents/rlmagents/__init__.py, libs/rlmagents/deepagents_cli/main.py, libs/deepagents/deepagents/__init__.py, libs/harbor/deepagents_harbor/__init__.py.
What it can do
Execute multi-agent reinforcement learning workflows
Workflow configuration and agent parameters → Coordinated agent actions and workflow results
Train reinforcement learning agents
Training data and environment parameters → Trained RL agent models
Orchestrate multiple AI agents
Agent definitions and task specifications → Coordinated multi-agent system execution
Run reinforcement learning simulations
Environment setup and agent policies → Simulation results and performance metrics
Manage agent interactions and communication
Agent communication protocols and message formats → Structured inter-agent communications
Deploy reinforcement learning workflows
Trained models and deployment configuration → Running multi-agent RL system
Tags
Tech Stack
Comments (0)
No comments yet
Editorially curated, with community endorsements as a secondary signal. Corrections welcome.