# LazyLLM

> Low-code Python framework for building, deploying, and fine-tuning multi-agent LLM applications.

LazyLLM is an open-source, low-code development tool from LazyAGI for building multi-agent large language model applications. It covers application construction, data preparation, model deployment, fine-tuning, and evaluation. The latest release listed is v1.4.0a0, and the project is licensed under Apache 2.0.

## What It Is

LazyLLM is a Python framework in which developers assemble AI applications from modules, components, and flows. Its documented workflow runs from prototype building to data feedback to iterative optimization. Developers build a prototype, analyze bad cases with task-specific data, then iterate on algorithms and fine-tune models.

## How Applications Are Built

Flows such as pipeline, parallel, diverter, if, switch, and loop define how data passes between callable objects. Modules such as OnlineChatModule, TrainableModule, ServerModule, UrlModule, and WebModule add training, deployment, inference, and evaluation. A built-in WebModule serves a chat interface, and the CLI can start a chatbot (`lazyllm run chatbot`) or a RAG bot (`lazyllm run rag`).

## Models, RAG, and Deployment

The project offers one interface for online model providers and locally deployed models. The docs list platforms including OpenAI, Qwen, Kimi, Zhipu, Doubao, SiliconFlow, MiniMax, and Nova. Local inference and fine-tuning frameworks include vLLM, LMDeploy, LightLLM, LLaMA-Factory, Collie, and Alpaca-LoRA. RAG components include Document, Retriever, Reranker, and node-group transforms such as sentence splitting. Applications can be deployed with a lightweight gateway, and images can be packaged for Kubernetes.

## Setup Path

Install with `pip3 install lazyllm`, or use the lazyllm/lazyllm Docker image or a source install. Run `lazyllm install standard` or `lazyllm install full` for extra dependency groups. The docs include a Cookbook of examples and a RAG tutorial series.

## Features
- Low-code multi-agent application assembly
- Flows: pipeline, parallel, diverter, if, switch, loop
- One-click deployment and image packaging
- Unified interface for online and local models
- Model fine-tuning with automatic framework selection
- RAG components: Document, Retriever, Reranker
- Built-in web chat interface (WebModule)
- CLI for chatbot and RAG launch
- Cross-platform launchers for bare metal and Slurm

## Integrations
OpenAI, Qwen, Kimi, Zhipu, Doubao, SiliconFlow, MiniMax, SenseNova, AI Ping, vLLM, LMDeploy, LightLLM, LLaMA-Factory, Collie, Alpaca-LoRA, Infinity, Kubernetes, Slurm, MCP, Docker

## Platforms
WINDOWS, MACOS, LINUX, API, DEVELOPER_SDK, CLI

## Pricing
Open Source

## Version
v1.4.0a0

## Links
- Website: https://docs.lazyllm.ai/
- Documentation: https://docs.lazyllm.ai/en/stable/
- Repository: https://github.com/LazyAGI/LazyLLM
- EveryDev.ai: https://www.everydev.ai/tools/lazyllm