Lang-Agent is a programmable AI agent configuration platform built on LangGraph, designed for developers needing precise control over complex agent workflows. It offers a visual, ComfyUI-inspired interface for assembling agents from nodes, differentiating itself by allowing custom state variables that influence node and conditional edge execution, enabling more sophisticated logic than traditional workflow tools.

How It Works

The platform leverages LangGraph for its core agent execution engine. Its key innovation lies in the support for custom state variables, which act as a shared, mutable dictionary accessible by all nodes and conditional edges within an agent. This allows for fine-grained control over data flow and decision-making, moving beyond simple sequential input/output chaining. The architecture comprises a FastAPI backend and a React-based frontend utilizing ReactFlow for the visual canvas, promoting extensibility through custom node development.

Quick Start & Requirements

Backend installation involves cloning the repository, setting up a Python environment with Poetry (poetry env use python, poetry shell, poetry install), and running the main module (python -m lang_agent.main). Frontend setup requires Node.js and Yarn, followed by dependency installation (yarn install) and launching the development server (yarn dev). The default access address is http://localhost:8820.

Highlighted Details

• Visual Agent Builder: A drag-and-drop interface (ReactFlow) for designing agent logic using nodes and edges.
• Custom State Management: Extensible state variables beyond the default messages for complex data handling and conditional branching.
• Extensible Node Ecosystem: Includes built-in nodes for LLM interactions, vector storage/retrieval, code execution, document handling, and user input.
• Pre-built Agent Templates: Offers ReactAgent for tool usage via MCP adapters and SupervisorAgent for orchestrating other agents.
• Configuration Hub: Centralized UI for managing LLM connections (OpenAI compatible), vector databases (Postgres, Milvus), and MCP tools.

Maintenance & Community

The provided README does not contain specific details regarding maintainers, community channels (like Discord or Slack), or project roadmaps.

Licensing & Compatibility

The project is licensed under the Apache-2.0 License, a permissive license that allows for commercial use and integration into closed-source projects.

Limitations & Caveats

Currently, LLM channel support is limited to OpenAI-compatible APIs, and vector database integrations are restricted to Postgres and Milvus. Custom state variables do not support list types. The platform encourages custom development, suggesting a focus on power users and flexibility over immediate out-of-the-box simplicity.