AI agent backend that treats your app as the real client

What it does

Cheshire Cat AI is a Python framework that runs custom LLM agents as a containerized microservice. It exposes WebSocket chat and a customizable REST API, letting your own application handle the frontend while the cat handles reasoning, memory, and tool use. It comes with built-in RAG via Qdrant, multiuser support with granular permissions, and an admin panel for management.

The interesting bit

Rather than offering another chat wrapper, it acts as backend infrastructure: you bring the model via LangChain, Qdrant handles the memory, and you extend behavior through a “Mad Hatter” plugin system that lets you inject logic at specific events or register function-calling tools with decorators.

Key highlights

• Ships as a Docker container with Qdrant included for vector search and memory
• Plugin architecture with event hooks (@hook), tools (@tool), and Pydantic-driven conversational forms (@form)
• Multiuser-ready with identity-provider agnostic permissions
• Admin panel and auto-generated API docs included out of the box
• GPL3 licensed with an explicit code of ethics and active Discord community

Caveats

• The project is in a transitional phase: version 2 is underway, and only bug-fix PRs are being accepted for v1
• Contributors should expect to discuss changes in an issue before submitting a PR; unsolicited features will not be merged

Verdict A solid fit for teams that need a self-hosted, extensible conversational backend without SaaS lock-in. Less appealing if you want a managed, turnkey chatbot platform where you never touch Python or Docker.