Agent Lightning
An open-source framework for training AI agents with reinforcement learning while keeping existing agent harnesses, tools, and workflows in the loop.
The Linkredibles take
Why it's worth knowing.
Train and optimize AI agents using real agent workflows without having to completely rewrite the agent itself.
Topics
Open source repository
Explore the source behind Agent Lightning.
Read the source, explore the project, and see how it is being developed on GitHub.
Keep exploring
You might also like.
514 Labs Agent Skills
A collection of Agent Skills following the open specification, including experimental workflows for building and evaluating software-agent variants.
555cider Agent Skills
A collection of Agent Skills shared across coding-agent harnesses such as Claude Code and Codex, following the SKILL.md directory convention.
A-Tokyo Agent Skills
Reusable Agent Skills for AI coding assistants including production engineering, development quality, and workflow automation.