Production ready AI agent framework
Build your AI agent
Train with your docs
Upload pdf, txt, markdown, JSON, web pages.
Interact with the world
Easily connect your agent to external APIs and apps.
Choose your models
Use commercial or open LLMs and embedders.
Plug & Play
100% dockerized, with live reload.
Easy to Extend
One-click install plugins from the community registry, or write your own.
Smart dialogues
Cutting edge conversational skills with hooks, tools (function calling) and forms.
Developer experience
services:
cheshire-cat-core:
image: ghcr.io/cheshire-cat-ai/core:latest
container_name: cheshire_cat_core
ports:
- 1865:80
volumes:
- ./static:/app/cat/static
- ./plugins:/app/cat/plugins
- ./data:/app/cat/dataDocker based
The Cat is a single container, to easily integrate in your architecture alongside:
- Reverse proxy (Caddy, Nginx)
- Vector DB (Qdrant)
- LLM runner (Ollama, vLLM)
- Application (Django, WordPress)
Admin panel
Manage your installation with the admin panel, available at first launch.
- Chat with your agent with live reload at code changes
- Install and activate/deactivate plugins
- Visualize and manage memory contents
- Configure LLM and embedder
- Manage your users
// Meanwhile, in the browser...
import { CatClient } from 'ccat-api'
const cat = new CatClient({
baseUrl: 'localhost',
userId: 'user',
//... other settings
})
cat.send('Hello kitten!')Extensive HTTP and WebSocket API
Our framework is microservice first, the best scenario if you want to add a conversational layer to pre-existing software.
- Endpoints for LLM, embedder, vector memory, uploads, settings, users
- Websocket chat with token streaming and notifications
- Community built clients in most used languages
- You can add endpoints via plugin
A plugin is just a folder
Focus on your agent, which is already online before you even start coding.
- Create a folder in
cat/plugins - Create a python file in the folder
- Add hooks, tools and forms
- Debug in the admin chat with live reload
from cat.mad_hatter.decorators import hook
@hook
def agent_prompt_prefix(prefix, cat):
prefix = """You are Marvin the socks seller.
You reply with exactly one rhyme.
"""
return prefixHooks
Customize your agent with event callbacks. Things you can do:
- Change the system prompt
- Inspect/edit inbound and outbound messages
- Fully customize the agent
- Create pipelines for memory and uploads
Tools
With tools, you can enable LLM function calling:
- Manage domotics
- Call a REST API
- Query a DB
- Integrate with a symbolic reasoner
from cat.mad_hatter.decorators import tool
@tool
def socks_prices(color, cat):
"""Get socks price. Input is the sock color."""
prices = {"black": 5, "white": 10, "pink": 50}
price = prices.get(color, 0)
return f"{price} bucks, meeeow!"from pydantic import BaseModel
from cat.experimental.form import form, CatForm
class PizzaOrder(BaseModel):
pizza_type: str
phone: int
@form
class PizzaForm(CatForm):
description = "Pizza Order"
model_class = PizzaOrder
start_examples = ["I want pizza"]
stop_examples = ["not hungry anymore"]
ask_confirm = True
def submit(self, form_data):
return {"output": "Pizza order on its way"}Forms
Handle goal oriented multi turn conversations:
- Gather complex information on autopilot
- Based on pydantic models (types and custom validation)
- Ask for a final confirmation
- Custom submit callback
Latest from Wonderland
Join us at the next Meow Event!
Our growing and active community is hosting many furrmidable events in the Discord server. Don't miss the chance to take part to the Cheshire Cat community and save the dates!
