LcStack is a configuration framework for LLM components, designed to quickly build your own LLM Agent and Workflow in a LowCode manner. Built on LangChain and LangGraph, it provides a user-friendly configuration environment and component library to manage the configuration process and Workflow, as well as the ability to integrate with other frameworks.

• Compatible with existing LangChain and LangGraph libraries and frameworks.
• Includes standard Chains and Agents, with options to add new ones or integrate with other frameworks.
• Easily set up Agents with essential components like LLM, PromptTemplate, Embedding, etc.
• Design Workflows using existing Agents and Workflows.
• Re-organize input/output to match upstream and downstream nodes.
• Offers a command-line tool for fast testing.

pip install -U lcstack

• Basic usage: basic.ipynb
• Common components: components.ipynb
• Workflow: workflow.ipynb
• Workflow (hierarchy): workflow_hierarchy.ipynb

• Basic components
  • LLM / Chat Model
  • PromptTemplate
  • Embedding
  • VectorStore
  • Memory / Chat History
  • Tools
• LangChain Chains
  • LLM Chain
  • Retrieval Chain
  • Document QA
  • LLM Requests Chain
  • SQL Query Chain
  • Router Chain
  • Documents Load Pipeline
  • (More coming soon)
• LangChain Agents
  • ReAct
  • Supervisor
  • Plan-and-Execute
• Workflow
  • Node2Node (for single-node processing of non-conditional nodes)
    • Node connection
    • Conditional branch processing and state modification
    • Manually adding execution nodes (not recommended)
  • TODO: Node2Nodes (parallel processing of non-conditional nodes)

• cli.py

# run the llm_chain in llm_chain.yaml
python cli.py -c llm_chain.yaml -i llm_chain -q "what's your name?"

# with intermediate output
python cli.py -c llm_chain.yaml -i llm_chain -q "what's your name?" -v

# run the wf_hierarchical_team in graph_supervisor_header.yaml, `query` as input key, and generate graph image
python cli.py -c wf_hierarchical_team/graph_supervisor_header.yaml -i teams_manager -q "tell me three jokes first, one about dog and two about apple, but only one at a time; and then teach me 2 knowledges, one about dog, and one about apple" -I query -G teams_manager.png

# for help
python cli.py --help