Cheshire Cat AI
Cheshire Cat AI provides a pure Python framework for building custom AI agents with minimal setup. Developers can install the package with uv , run a single command, and start building agents using four core primitives: Agent, tool, Directive, and lifecycle events. The framework supports type-hinted tool definitions for automatic LLM function calling and directive-based middleware for RAG, memory, and guardrails.
- Artificial Intelligence
- AI Agents
- Developer Tools
- Software Only
Funding
Founders
Founder details are not available yet.
Product
Problem
Building custom AI agents typically requires complex orchestration of multiple libraries, APIs, and infrastructure components. Developers face steep learning curves and significant boilerplate code when trying to implement agent loops, tool calling, memory, and middleware, which slows down experimentation and iteration.
Solution
cheshirecat.ai offers a pure Python framework that simplifies AI agent development through four core primitives: Agent, tool, Directive, and lifecycle events. Developers can install the package with `uv`, run a single command, and immediately start building agents that read chat, call tools, and generate responses. The framework uses type-hinted, docstringed methods for tool definitions, allowing LLMs to invoke functions via standard function calling. Middleware such as RAG, memory, skills, and guardrails are implemented as directives that wrap the agent loop, while lifecycle events enable reactive behavior with simple one-argument functions. This design keeps the system small, composable, and easy to reason about, letting developers start with a basic agent and progressively open it up to more advanced capabilities.
Target Audience
Primary users are Python developers, AI hobbyists, and technical teams who want to learn, prototype, and build custom AI agents with minimal setup and maximum control over the agent loop.
Features
- Single-command installation and startup using `uv` and the `ccat` command, with a local web interface at http://localhost:1865
- Four core primitives (Agent, tool, Directive, lifecycle events) importable from a single `from cat import ...` statement
- Type-hinted and docstringed tool methods that enable automatic LLM function calling
- Directive-based middleware system for implementing RAG, memory, skills, and guardrails
- Lifecycle event hooks that accept one-argument functions for reacting to agent state changes
- Pure Python implementation with no external service dependencies for local development