Cheshire Cat AI (Stregatto) is an open-source framework and microservice for building AI agents and conversational applications, designed from the ground up to be easy to understand and easy to extend. Maintained by the cheshire-cat-ai community on GitHub, where it has about 3,100 stars, it is actively developed in 2026 and describes itself as the best framework to learn how AI agents work, with an education and research focus alongside real production users.
The Cat runs as a Dockerised microservice exposing REST APIs and a built-in admin interface. Under the hood it combines an LLM of your choice with declarative and episodic memory stored in a vector database, a plugin system for adding tools and hooks, function calling, and MCP client support for connecting external tool servers. Because the internals are deliberately readable, developers can follow the full agent loop (input processing, memory recall, tool execution, response generation) and then customise each stage with their own code.
Around the core sits a community plugin ecosystem: developers publish plugins that add tools, integrations and behaviours, and Cat instances can install them from within the admin panel. The framework supports streaming responses and real-time interaction, and its memory system distinguishes between conversation history, facts extracted from dialogue and documents you upload for retrieval. The current generation of the codebase is a ground-up rewrite aimed at agentic engineering workflows, so teams evaluating it should expect a fast-moving project with an engaged community behind it.
Cheshire Cat AI is free and open source under the GPL-3.0 license. You pay only for your own hosting and any LLM APIs you connect.
Developers, students and research groups who want a transparent, hackable agent framework for building and learning about conversational AI systems.