rllm
rLLM is an open-source framework for post-training language agents via reinforcement learning. You build custom agents and environments, train them with reinforcement learning and deploy them for real workloads. The README's release notes mention v0.2.1 (December 2025) with support for the Tinker backend, LoRA and vision-language model training and the Eval Protocol, v0.2 (October 2025) which introduced an AgentWorkflowEngine for training over arbitrary agentic programs, and open models released with the project such as DeepSWE-Preview, a 32B software-engineering agent, and DeepCoder-14B-Preview, a 14B coding model.
It requires Python 3.10 or newer (3.11 if you use the Tinker backend) and can be installed with pip or uv directly, built from source, or run with Docker; the verl or tinker backend is chosen at install time, for example `uv pip install "rllm[verl] @ git+https://github.com/rllm-org/rllm.git"`. The README also lists projects built with rLLM. The code is Apache-2.0 licensed.
Features
Why choose it
Trade-offs
Maintenance check
- Last commit
- 2026-10-09
- Commits, last 90 days
- 7
- Releases, last 90 days
- 0
- Latest release
- v0.3.0-pre (2026-04-30)
- Open issues
- 172
- Recent issues with at least one comment
- 5/10
Oldest week on the left; the latest four weeks are orange. Source: GitHub commit activity.
Status rule: archived on GitHub = Archived; last commit within 90 days = Active; 91 to 365 days = Slowing; older = Stale. All figures come from the GitHub API on the date shown.
Most discussed open issues
- #30 oom when update_policy15 comments · 2025-02-15
- #434 Qwen3.5 support13 comments · 2026-03-10
- #29 Extremely Long Execution Time per Step12 comments · 2025-02-15
- #33 Reproduce deepscaler10 comments · 2025-02-17
- #7 great job！have some questions to ask10 comments · 2025-02-12
Issue titles are quoted from GitHub. Open the link to check whether a problem still applies to the version you use.
Compatibility
Quick start
Use cases
Alternatives
Comments
No comments yet. Be the first!