GH Repository · areal-project
AReaL
The RL Bridge for LLM-based Agent Applications. Made Simple & Flexible.
- stars
- 5,768
- 30-day movement
- +134/day
- Related entries
- 67
- Connections
- 4
python/uvpythondockerPythonllm-agentagentrlmlsysagent-appmachine-learning-systemsllm-reasoningreinforcement-learningllm
AReaL is a Python-based framework described as 'The RL Bridge for LLM-based Agent Applications', positioning itself as simple and flexible. It connects reinforcement learning training with LLM agent applications, and ships with agent configuration files (AGENTS.md, CLAUDE.md, .claude directory) plus a uv-based Python toolchain and Docker support.
You want to apply reinforcement learning to LLM-based agents through a single framework rather than wiring training and agent infrastructure yourself.
Use it to
- Train LLM agents with reinforcement learning
- Bridge RL training pipelines with agent applications
- Run experiments in the provided Docker/uv environment
- Explore RL for LLM reasoning tasks
For ML engineers and researchers working on RL for LLM agents
- Role
- agent-app
- Language
- Python
- Licence
- Apache-2.0
- Forks
- 603
- Open issues
- 19
- Last push
- 2026-09-17
- Latest release
- v0.1.1 · 2025-03-04
- Skills shipped
- 7
topicsreinforcement-learningllm-agentsllm-reasoningmlsyspythonagent-applications