Support
Support runs through GitHub Discussions and community organised Meow Events listed on the site. There is no commercial support tier mentioned on the homepage.
Learn moreCheshire Cat AI
A Python agent framework for people who learn by building
Intermediate API
Own Cheshire Cat AI? Verify ownership to get a dofollow link
What is Cheshire Cat AI?
Cheshire Cat AI is a developer tool. Cheshire Cat is a pure Python framework for building AI agents from four primitives: agents, tools, directives and hooks. It ships a local web UI, native MCP support, streaming over AG-UI and model swapping across OpenAI, Anthropic, Gemini, Ollama, OpenRouter and vLLM.
Cheshire Cat is a framework for AI agents written in pure Python. It is aimed at developers, learners and tinkerers who want to understand how an agent actually works rather than configure one through a hosted dashboard. The whole model rests on four primitives that all come from a single import. An Agent is a loop that reads the chat, calls tools and replies; you subclass it, give it a slug, a name and a system prompt, and it exists. A Tool is a method on that agent, type hinted and documented, which the model calls through function calling. A Directive is middleware wrapped around the agent loop, and the project treats retrieval, memory, skills and guardrails as directives too. A Hook is a one argument function that reacts to lifecycle events such as before and after an agent run. In practice the setup is deliberately small. Install the package with uv, run the ccat command, and a multi-chat, multi-agent web interface opens on localhost:1865. The first run walks you through picking a model. From there you either write plugins by hand or hand the repository to a coding agent and describe what you want. Plugins are folders: drop one in and it can contribute agents, tools, directives and endpoints without boilerplate. Model choice is a one line swap across OpenAI, Anthropic, Gemini, Ollama, OpenRouter and vLLM, and the framework speaks the Model Context Protocol natively, so external servers bring their tools, prompts and resources along. Projects that outgrow a single assistant can run several agents in one conversation, picking one per message or letting them call each other. Everything streams in real time over the AG-UI protocol, covering tokens, tool calls and lifecycle events. The HTTP layer is part of the same package: a decorator exposes a route, and role based authentication is built in, so custom endpoints and models do not need extra wiring. Messages reach an agent at a predictable REST path, which makes it straightforward to put your own front end in front of the Cat. Compared with visual agent builders, Cheshire Cat assumes you are comfortable in an editor and would rather read source than click through menus. It sits closest to other code first orchestration libraries, but bundles a running server, a UI and a plugin layout so a working agent is minutes away. The community gathers in GitHub Discussions and at Meow Events.
How do you use Cheshire Cat AI?
1Install the package with uv
Initialise a bare project, add the cheshire-cat-ai package and run the ccat command. The server starts locally and points you at the web interface.
2Pick a model on first run
The interface asks which model provider to use. Choose a hosted API or a local Ollama instance, and the agent is ready to answer.3Write your first agent
Subclass Agent, set the slug, name and system prompt, then reload. The new agent appears in the chat interface alongside the default one.4Attach a tool with function calling
Decorate a method with the tool decorator, add type hints and a docstring. The model reads the signature and calls the method when the conversation calls for it.5Add a directive for shared behaviour
Wrap reusable logic such as time injection or guardrails in a Directive and list it on any agent that needs it. Directives run as middleware around the agent loop.6Follow a worked example on the blog
Read a full build write up, such as hooking a fine tuned model into the Cat, before extending the project further.
Pros and cons
Pros
- The four primitive model (agent, tool, directive, hook) is learnable in an afternoon and covers most agent patternsAI
- Local setup is a single install and run command, with a working multi-agent web UI on first launchAI
- Model agnostic by design, so the same agent can run against hosted APIs or a local Ollama instanceAI
- Native MCP support lets external servers contribute tools, prompts and resources without custom glueAI
- Plugins are plain folders, which keeps extension work free of registration boilerplateAI
Cons
- Requires Python and command line comfort; there is no hosted sign up path for non-developersAI
- The homepage gives no pricing, licensing tiers or hosted offering, so cost of running it falls on your own infrastructureAI
- Documentation lives on a separate docs site, so newcomers juggle two sources while learningAI
- Ecosystem and third party tutorials are smaller than those around mainstream agent SDKs, which means fewer worked examplesAI
- The framework is evolving quickly, so code samples from older releases may not match current APIsAI
How much does Cheshire Cat AI cost?
Integrations
Model providers listed on the homepage are OpenAI, Anthropic, Gemini, Ollama, OpenRouter and vLLM. Model Context Protocol servers can be connected natively so their tools, prompts and resources are available to agents.
Features
Agents, tools, directives and hooks make up the core model, all importable from one module. Multi-agent chats let several agents share a conversation. Model providers include OpenAI, Anthropic, Gemini, Ollama, OpenRouter and vLLM, swappable with one line. Native MCP support, real time streaming over AG-UI, folder based plugins, custom endpoints with built-in role based auth, and a local multi-chat web UI round out the feature set.
Frequently asked questions about Cheshire Cat AI
Does Cheshire Cat AI have an API?
Yes, Cheshire Cat AI offers an API.How do you use Cheshire Cat AI?
The walkthrough on this page covers 6 steps: 1. Install the package with uv 2. Pick a model on first run 3. Write your first agent 4. Attach a tool with function calling 5. Add a directive for shared behaviour 6. Follow a worked example on the blog.What platforms does Cheshire Cat AI support?
Cheshire Cat AI is available on Web App, Linux, MacOS and Windows.What does Cheshire Cat AI integrate with?
Model providers listed on the homepage are OpenAI, Anthropic, Gemini, Ollama, OpenRouter and vLLM. Model Context Protocol servers can be connected natively so their tools, prompts and resources are available to agents.What are the limitations of Cheshire Cat AI?
Requires Python and command line comfort; there is no hosted sign up path for non-developers. The homepage gives no pricing, licensing tiers or hosted offering, so cost of running it falls on your own infrastructure. Documentation lives on a separate docs site, so newcomers juggle two sources while learning.
