Cheshire Cat AI

Named with a nod to Lewis Carroll's famous character, the framework is celebrated in the developer community for its flexibility, extensibility, and freedom from vendor lock-in developers can use any language model, any embedding model, and any infrastructure they prefer.

The framework's Apache 2.0 license and active open-source community make it a compelling foundation for organizations building AI-powered products.

At its technical core, Cheshire Cat implements retrieval-augmented generation (RAG) not only on documents and knowledge bases but also on conversation history a crucial distinction that enables the AI to draw on the full context of past interactions rather than just its static training data and uploaded files.

This conversational RAG capability makes the AI genuinely contextually aware across extended interactions, remembering relevant information from earlier in the conversation to inform current responses. The framework uses Qdrant as its vector database for semantic search and knowledge retrieval.

The plugin architecture is central to Cheshire Cat's extensibility philosophy.

Developers extend the framework's capabilities by writing plugins that add custom tools (function calling), hooks (event callbacks that modify behavior at specific pipeline stages), and conversational forms (structured multi-step data collection flows).

The plugin marketplace allows developers to share and install community-built extensions, creating a growing ecosystem of specialized capabilities built on the core framework. This plugin-driven model keeps the core framework lean while supporting unlimited functional expansion.