OpenPipe/ART

Agent Reinforcement Trainer: train multi-step agents for real-world tasks using GRPO. Give your agents on-the-job training. Reinforcement learning for Qwen3.6, GPT-OSS, Llama, and more!

10.8k stars · 1k forks · Python

Healthy

Healthy: actively maintained by 4 people, no known risks.

Why

No rule fired.

What would change this

Maintenance

Partial

Active

Human commits, 30 days
149
Human commits, 90 days
244
Human commits, 365 days
527
Bot share of commits, 365 days
3%
Last commit
Oct 12, 2026 (2 days ago)
Last human commit
Oct 12, 2026 (2 days ago)
Authors, 90 days
4
Last release
Aug 14, 2026 (2 months ago)
Median PR merge time
1 hour

Responsiveness: of 0 recent issues, n/a got a maintainer reply. Median first response: n/a.

Sources: Commits, Releases, Merged pull requests, Issues

People

Complete
Owner
Organization
Commits sampled
486 since Oct 14, 2025
Authors for half of commits
1
Authors for 80% of commits
3
Contributors
25

Top committers

Sources: Commits, Contributors

Stars

Complete

Organic, confidence medium

Stars per week, 2025-03-09 to 2026-10-11. 20 weeks overlap a burst.Peak 1.8k a week2025-03-092026-10-11
Bars show stars gained each week. Orange bars overlap a burst.

Bursts

Anyone can buy stars for any repo; bursts alone don't prove the owner did.

Sources: How star history is analyzed

Risk

Complete

Flags

No risk flags.

License

Now
Apache-2.0 (OSI-approved)
A year ago
Apache-2.0 (OSI-approved)

Packages

Package and advisory data: deps.dev and OSV.

Sources: GitHub: OpenPipe/ART, deps.dev: PyPI openpipe-art

Badge

Add this to a README to show the current verdict:

[![RepoRadar](https://reporadar.ai/OpenPipe/ART/badge.svg)](https://reporadar.ai/OpenPipe/ART)

Earlier verdicts

Computed 7 hours ago · method 2026.10.2