ai-agent

agent-lightning

Name: agent-lightning
Author: microsoft

The absolute trainer to light up AI agents.

18,872 stars1,701 forksPythonUpdated 11/2/2026100% free · open source

What it does

Agent‑Lightning lets you quickly prototype, train, and evaluate autonomous LLM‑based agents with a few CLI commands.

When to use it

•You need a self‑contained way to turn a prompt‑based workflow into a loop‑back agent that can call tools and persist state.
•You want to benchmark different prompting strategies (react‑loop, tool‑use, memory) without writing custom RL code.
•Your startup is building a customer‑support bot that must plan, execute, and self‑correct on real‑time data.

Ready-to-paste prompt

agent-lightning run --config examples/travel_agent.yaml --input "Schedule a 2‑hour video call with the product team tomorrow at 10 AM PST and add it to my Google Calendar."

Heads up: The CLI validates the presence of an LLM API key at runtime; if `OPENAI_API_KEY` (or `AZURE_OPENAI_ENDPOINT`/`AZURE_OPENAI_KEY`) is missing the command exits with a cryptic error. Set the env var before the first run.

Saves to your device

Use with Claude

New

Skip the builder — one click puts this in Claude, Cursor, Antigravity and more.

🔑 Needs an API key

Works after you add credentials — the setup agent will ask you for them.

Try it instantly — no install

Claude Code

mkdir -p ~/.claude/skills/agent-lightning && curl -fsSL https://workflowstacks.com/api/skills/agent-lightning/claude-skill -o ~/.claude/skills/agent-lightning/SKILL.md

Open in another AI app

Opens the app with this repo with the prompt ready to go — no copy-paste needed.

Connect the whole catalog (MCP)

claude mcp add --transport http workflowstacks https://workflowstacks.com/api/mcp

Adds a WorkflowStacks connector to Claude Code: search and load any skill here by chatting.

How agent-lightning works

Codeflow

Free to inspect

Agent-lightning is a medium Python project (~12k lines across 86 code files, plus 24 test files). Setup is light: installs like a normal app. Reading the code is optional. Last commit this month, MIT license, has a test suite.

Size

Medium codebase

~12k lines · 86 code files · ~2 h to skim

Setup

One-command install

Installs like a normal app. Reading the code is optional.

Runs on

Python

No API keys detected

Python 99%Shell 1%

Where to start reading

1
README.md
Start here — what it does and how to install it
2
pyproject.toml
Dependencies and the commands it exposes
3
agentlightning/__init__.py
Inside agentlightning/ — the main logic begins here
4
examples/calc_x/.dockerignore
A worked example — copy this to get going

What's in each folder

agentlightning/Core code — the actual logic38 files

skills/Prompts, skills & agent definitions9 files

examples/Examples you can copy68 files

docs/Documentation35 files

scripts/Helper scripts3 files

tests/Tests — proof it works24 files

.github/CI / automation (GitHub Actions)6 files

.agents/Tool / agent settings2 files

READMEHas testsDocumentedCI checksExamples includedMIT licenseUpdated this month