MaxKB

MaxKB is model-agnostic and supports a wide range of private and public large language models, including DeepSeek, Llama, Qwen, OpenAI, Claude, Gemini, MiniMax, and models served via Ollama or Xinference. Any provider exposing a standard OpenAI-format API can also be integrated with a custom Base URL.

MaxKB processes uploaded documents or web-crawled content through automatic chunking and vectorization, storing embeddings in a pgvector-backed PostgreSQL database. At query time, multi-route recall retrieves the most relevant segments and injects them as grounded context into the LLM prompt, significantly reducing fabricated responses compared to vanilla chat completions.

Yes. MaxKB provides embeddable dialogue widgets and native connectors for enterprise messaging platforms like DingTalk, Lark (Feishu), WeChat Work, and Slack (Pro Edition). For deeper integration, an Open API is available in the Pro Edition, enabling programmatic access to agents and knowledge base queries from external systems.

A Simple Agent in MaxKB provides a streamlined Q&A interface backed by a knowledge base with minimal configuration. An Advanced Agent unlocks the full visual workflow canvas, supporting multi-node orchestration with conditional branching, loop nodes, variable assignment, sub-agent calls, MCP tool-use, and form collection — suitable for automating complex, multi-step business processes.

Absolutely. MaxKB is designed for private deployment via Docker. It integrates natively with local model runners like Ollama and Xinference, meaning organizations can run DeepSeek, Llama, or Qwen models entirely on their own infrastructure with no data leaving the premises — a critical requirement for regulated industries.