Skip to content

Examples Catalog

Want to Contribute?

We welcome contributions to the examples catalog! Please refer to the Contributing guide for more details.

  • APO room selector

    Prompt-optimize a room-booking agent with the built-in APO algorithm, then contrast it with the write-your-own algorithm and debugging workflows in the tutorials. Pairs well with the Train the First Agent how-to and the Write the First Algorithm guide.

    Browse source

  • Azure OpenAI SFT

    Run a supervised fine-tuning loop against Azure OpenAI: roll out the capital-lookup agent, turn traces into JSONL, launch fine-tunes, and redeploy the resulting checkpoints through Azure CLI.

    Browse source

  • Calc-X VERL math

    VERL-based reinforcement learning setup for a math-reasoning agent that uses AutoGen plus an MCP calculator tool to solve Calc-X problems end to end.

    Browse source

  • Claude Code SWE-bench

    Instrumented driver that runs Anthropic's Claude Code workflow on SWE-bench instances while streaming traces through Agent-lightning—supports hosted vLLM, official Anthropic, or any OpenAI-compatible backend and emits datasets for downstream tuning.

    Browse source

  • Minimal building blocks

    Bite-sized scripts that isolate Agent-lightning primitives (e.g., LightningStore usage, LLM proxying, minimal vLLM host) so you can study each part before composing larger workflows.

    Browse source

  • RAG (MuSiQue)

    Retrieval-Augmented Generation pipeline that preps a Wikipedia retriever via MCP and trains a MuSiQue QA agent with GRPO. Documented for historical reference (verified on Agent-lightning v0.1.x).

    Browse source

  • Search-R1 RL

    Reproduction of the Search-R1 workflow that prepares its own retrieval backend, runs the rollout script, and coordinates GRPO-style training without extra orchestration layers (last validated on v0.1.x).

    Browse source

  • Spider SQL agent

    LangGraph-powered text-to-SQL workflow for the Spider benchmark, combining LangChain tooling with Agent-lightning rollouts; follow along with the how-to for training SQL agents.

    Browse source

  • Tinker integration

    Adapter package (agl_tinker) with Tinker plus sample CrewAI/OpenAI agents that feed Agent-lightning traces into Tinker’s reinforcement-learning backend for both toy and 20-Questions-style workflows.

    Browse source

  • Unsloth SFT

    Supervised fine-tuning loop that ranks math-agent rollouts, fine-tunes with Unsloth’s 4-bit LoRA stack, and mirrors the Fine-tune with Unsloth recipe.

    Browse source