🛠️Agent Lightning
microsoft/agent-lightning · homepage ↗
The absolute trainer to light up AI agents.
⭐ Very popular: 17k stars, gaining about 17 a week
View on GitHub ↗repo profile
momentum
durability
since we covered it
why it's a big deal
- It lets teams train and optimize existing AI agents without rewriting them, by wrapping current code in lightweight helper calls instead of porting to a new framework.
- It works across LangChain, OpenAI Agent SDK, AutoGen, CrewAI, Microsoft Agent Framework, or plain Python OpenAI calls, so the training layer is not tied to one agent stack.
- It supports reinforcement learning, automatic prompt optimization, and supervised fine-tuning, and can optimize individual agents inside a multi-agent system.
under the hood
- It is primarily Python, with a TypeScript dashboard UI making up roughly 15 percent of the codebase.
- A tracer collects prompts, tool calls, and rewards as structured spans that flow into a central LightningStore, which keeps tasks, resources, and traces in sync.
- A Trainer streams datasets and moves resources between the store and the chosen algorithm, which reads spans and posts updated resources such as refined prompt templates or new policy weights back to the inference engine.
Radar summary, generated from the project's public sources
star history
- PR#20 3k 2025-10-29
- now 17k + 15k since first covered
curve is sampled from GitHub's star history, plus our own daily readings since we covered it; the dashed stretch is before we first covered it, the solid line since. figures at coverage are the numbers we printed then (approx.), current count is live.
understory
Better known than its recent output, coasting a little on attention.
- output, commits & releases
- clout, star velocity
output = commits & releases; clout = star velocity, both 0 to 100 monthly indices; the gap where output runs above clout is the understory. The understory →
covered in
-
Train any AI agent with reinforcement learning
similar projects
compare these →- 📖 ai-agent-book
2.2× the stars
AI Agent PDF
38k ACTIVE - 🧠 MaxKB
GPL-3.0 licensed
🔥 MaxKB is an open-source platform for building enterprise-grade agents.
23k ACTIVE - 🎨 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!
11k ACTIVE
comments
Sign in with GitHub to add your blip on Agent Lightning.