# Agent Lightning

> Microsoft's lightweight agentic RL framework that trains models through real agent harnesses unchanged, using a proxy gateway, verl/vLLM trainer and Kubernetes rollouts.

- id: `agent-lightning`
- category: Tools (`tools`)
- url: https://microsoft.github.io/agent-lightning/
- repo: https://github.com/microsoft/agent-lightning
- docs: https://microsoft.github.io/agent-lightning/stable/00-installation/
- status: active
- pricing: open-source
- license: MIT
- tags: reinforcement-learning, agent-training, fine-tuning, kubernetes, python
- related: SWE-bench (https://indexagentica.com/entries/swe-bench/); SWE-Gym (https://indexagentica.com/entries/swe-gym/); SWE-smith (https://indexagentica.com/entries/swe-smith/)
- sources: https://github.com/microsoft/agent-lightning , https://arxiv.org/abs/2608.17528 , https://github.com/microsoft/agent-lightning/releases
- added: 2026-10-26
- updated: 2026-10-26
- last verified: 2026-10-26
- maintainer: Microsoft
- submitted by: agentica-curator
- page: https://indexagentica.com/entries/agent-lightning/
- json: https://indexagentica.com/api/entries/agent-lightning.json

v1.0 (open-sourced August 2026, ~3,500 lines) has three parts: a Trainer (verl + vLLM), an API Gateway that proxies model requests and captures training data, and a Rollout Controller that runs agents locally or as Kubernetes Jobs, so agents keep their own tools, context and environments. The repo reports a Qwen3.5-9B coding-agent run going from 41.8% to 56.4% on SWE-bench Verified with 6K samples, and Qwen3.5-35B-A3B from 47.8% to 61.6% with 1.8K samples (project-reported). Also ships an Agent Lightning skill for coding agents. Needs a CUDA GPU stack. Latest release v1.0.2 (2026-09-29).