cqzyys/lang-agent

What it solves

Lang-Agent is a visual configuration platform for building AI Agents. It addresses the limitation of traditional workflow tools that only pass output from one node to the next by introducing customizable state variables. This allows for more precise control over the logic and data flow within an Agent, making it more flexible for complex business needs.

How it works

Built on top of LangGraph, the platform uses a drag-and-drop canvas where users can combine various nodes and edges to define an Agent's behavior.

• Nodes: These are the basic building blocks of logic. They include LLM nodes for text generation, VLM nodes for multimodal processing, and utility nodes like counters, transformers, and code executors.
• Edges: These define the execution order. Default edges follow a linear path, while conditional edges use state variables to decide which node to execute next.
• State Variables: A global dictionary within the Agent that stores data (like conversation messages) which can be accessed and updated by any node or conditional edge.
• Agents: The system supports pre-built agents like ReactAgent (for tool use via MCP) and SupervisorAgent (for coordinating other reusable agents).

Who it’s for

• Developers who want to visually design complex AI agents without writing boilerplate code.
• Users who need a high degree of control over the Agent's state and logic flow.
• Teams looking to build extensible AI workflows that allow for the creation of custom nodes.

Highlights

• State-Driven Logic: Uses global state variables to control node inputs and conditional branching.
• Extensible Architecture: Provides a clear path for developers to add their own custom nodes via frontend (React) and backend (Python) implementations.
• SaaS-like Configuration: Includes built-in management for model connections, MCP tools, and vector databases (Postgres, Milvus).
• Multimodal Support: Includes dedicated nodes for image loading and VLM processing.
• Integrated RAG: Features vector database configuration and document management for retrieval-augmented generation.