All 92 Dependencies
Every package Agent Lightning depends on, ranked by repo health score.
Agent Lightning tackles a specific problem in agent development: once you have a working agent built in LangChain, AutoGen, CrewAI, the OpenAI Agents SDK, or even plain Python, improving it usually means manual prompt tweaking with no systematic feedback loop. Agent Lightning instead wraps your existing agent and applies training algorithms — reinforcement learning, automatic prompt optimization, or supervised fine-tuning — to optimize its behavior, with the project's stated goal of requiring almost zero code changes to the agent itself.
It supports selectively optimizing one or more agents within a multi-agent system, rather than requiring an all-or-nothing training setup, and works regardless of which (if any) agent framework the target agent was built with. The project ships as the PyPI package `agentlightning`.
MIT licensed and maintained by Microsoft, Agent Lightning positions itself less as an agent framework and more as a training harness that sits alongside whatever framework you already use.
