Enterprise AI agents built on Django, LangChain, and pgvector

What it does

MaxKB is a web-based platform for building enterprise AI agents backed by internal documents. It combines a RAG pipeline—document upload, web crawling, text splitting, and vectorization—with a workflow engine that can call functions and MCP tools. Under the hood it uses Django, Vue.js, PostgreSQL with pgvector, and LangChain.

The interesting bit

Instead of chasing novel research architectures, MaxKB treats agent building as an integration problem. It supports private models like DeepSeek and Llama alongside public APIs such as OpenAI and Claude, and handles text, image, audio, and video natively. That breadth is unexpected for a stack this conventional.

Key highlights

• RAG pipeline with automatic document crawling, splitting, and vectorization to reduce hallucinations
• Visual workflow engine with a function library and MCP tool-use for orchestrating business logic
• Model-agnostic support for private and public LLMs, including multimodal input and output
• Zero-coding integration into existing third-party business systems
• GPL v3 licensed, distributed via Docker with offline install support

Caveats

• The README claims “zero-coding” integration but never explains the mechanism or API contract
• GPL v3 copyleft may complicate embedding in proprietary commercial products

Verdict

Organizations that need a self-hosted, document-backed agent suite with visual workflows should take a look. If you want a library to drop into an existing codebase, this is not it.