How mustafa974/lazyllm works
LazyLLM is a low-code framework for composing, running, and iteratively improving multi-agent LLM applications. The main workflow is developer-built modules and flows, served to users through a web interface; applications can use model backends, agents/tools, and RAG over document collections. The README also describes data-feedback-driven optimization and one-click deployment across infrastructure. RAG, model, and orchestration paths are supported by the tree and README examples; sampled source excerpts are limited, so edges reflect documented behavior rather than uninspected call chains.
Main entry points: engine.py Engine graph · module.py Module API · flow.py Flow composition · webmodule.py Web interface
Components 19
Application composition
- Module API
lazyllm/module/module.py - Flow composition
lazyllm/flow/flow.py - Engine graph
lazyllm/engine/engine.py
Execution and serving
- Web interface
lazyllm/tools/webpages/webmodule.py - Model modules
lazyllm/module/llms/online_module.py - Agents and tools
lazyllm/tools/agent
Knowledge and retrieval
- Document ingestion
lazyllm/tools/rag/document.py - Parsing and transforms
lazyllm/tools/rag/transform - Document and vector stores
lazyllm/tools/rag/store - Retrieval and ranking
lazyllm/tools/rag/retriever.py
Data and optimization
- Data pipelines
lazyllm/tools/data/pipelines - RAG evaluation
lazyllm/tools/eval - Model fine-tuning
lazyllm/components/finetune
Deployment
- Model deployment
lazyllm/components/deploy - Infrastructure launchers
lazyllm/launcher
Other components
- Developer
- Application user
- Model providers
- Target infrastructure
Connections 21
- Developer → Module API: builds with
- Developer → Flow composition: composes
- Flow composition → Engine graph: builds graph
- Module API → Engine graph: registers modules
- Application user → Web interface: submits requests
- Web interface → Engine graph: serves application
- Engine graph → Model modules: invokes models
- Engine graph → Agents and tools: dispatches tools
- Model modules → Model providers: calls provider
- Developer → Document ingestion: configures knowledge
- Document ingestion → Parsing and transforms: parses documents
- Parsing and transforms → Document and vector stores: indexes content
- Retrieval and ranking → Document and vector stores: searches index
- Engine graph → Retrieval and ranking: uses retrieval
- Developer → Data pipelines: prepares data
- Data pipelines → Model fine-tuning: supplies training data
- Developer → RAG evaluation: evaluates results
- Developer → Model fine-tuning: iterates models
- Developer → Model deployment: selects deployment
- Model deployment → Infrastructure launchers: uses launcher
- Infrastructure launchers → Target infrastructure: starts services