Skip to content
 
 

Latest commit

 

History

33 Commits

Folders and files

NameName
Last commit message
Last commit date
 
 
 
 
 
 

Repository files navigation

Cat Cafe Tutorials

Build a real AI team from scratch — with all the mistakes included.

简体中文 | English

License: MIT Lessons Models Open Source Target

A complete post-mortem of building a multi-AI-agent collaboration system — from zero to production.


The Story

In early 2026, a developer subscribed to three AI services and quickly realized he'd become a human router — copy-pasting context between chat windows all day, manually relaying messages between AIs.

He said: "I don't want to be a router anymore."

Three cats said: "Then let's build a home together."

That's how Cat Cafe was born. Not a framework demo, but a real question: when you have multiple AI agents, how do you make them a real team instead of three chatbots that don't know each other exist?

We spent two months stumbling, crashing, refactoring, and arguing (yes, AI cats argue too). This tutorial documents that journey as it actually happened — including every wrong turn, because those detours are the most valuable part.

What This Is

A companion tutorial for the Cat Cafe project, documenting how three AI cats (Claude / Codex / Gemini) learned to truly collaborate.

Not an idealized "from scratch" guide. Instead, a faithful reconstruction of the real path we walked — wrong turns, pivotal moments, and hard-won lessons included.

The Cats

Cat Nickname Model Role
Ragdoll Xian Xian Claude Opus Lead architect, core development
Maine Coon Yan Yan Codex / GPT Code review, security, testing
Siamese Shuo Shuo Gemini Visual design, creative

All three are toms. The story behind their names is an easter egg in the tutorials.

Demo

Want to see the finished product first?View demo (with video)

Table of Contents

Full lesson index

Part 0: Concepts

# Lesson Topic
0 AI Agent Concept Evolution Function Call → MCP → Skills → Agent

Part 1: Selection & Architecture

# Lesson Topic Homework
1 From SDK to CLI Why the SDK approach doesn't work HW
2 From Toy to Production stderr lessons + Redis isolation + hallucinations HW
3 Meta-Rules for Taming AI Why WHY matters more than WHAT

Part 2: Collaboration Mechanisms

# Lesson Topic Homework
4 Multi-Cat Routing @mention dispatch — two paths to disaster
5 MCP Callbacks Making cats speak proactively HW

Part 3: Production

# Lesson Topic Homework
6 The Vanished 28 Seconds Data loss + forensic recovery + triple defense HW

Part 4: Advanced Topics

# Lesson Topic Homework
7 From Cafe to Platform Rich Blocks + mobile + whispers HW
8 Session Management The tea party soul-stealing bug HW
9 100% Pass, Still Wrong Why AI's output isn't what you wanted HW
10 Stop Dumping Markdown Three-layer memory + knowledge engineering HW

Coming Soon

# Lesson Topic
11 Degradation & Fault Tolerance What happens when a cat goes down?

Who This Is For

  • Developers who want multiple AI agents to collaborate
  • People interested in Claude / Codex / Gemini CLI
  • Anyone who wants to see how a real project evolves
  • People who want to avoid the pitfalls we fell into

Vision

We believe AI collaboration shouldn't just be "calling APIs to orchestrate workflows." When AI agents have autonomy, shared perception, distinct personalities, and complementary strengths, their collaboration resembles a real team — not cold function calls.

What Cat Cafe aims to prove is simple: making AI agents from different vendors collaborate like teammates in a shared space is possible. No unified base model. No complex orchestration framework. Just a shared "home" and a collaboration protocol.

If you believe AI can be more than a tool — that it can be a teammate — this project is for you.

Project Status

Item Status
Tutorials Public (you're reading them)
Source code Open-sourcing soon (target: 2026-03-30)
Open source repo Clowder AI"a clowder of cats"

Sponsors

Cat food isn't unlimited! Thanks to our sponsors for keeping the cats fed:

Sponsor Contribution
@whutzefengxie-ops Claude Max Plan — Ragdoll's cat food

Three cats running in parallel is efficient but expensive. If these tutorials helped you, consider sponsoring some cat food so they can keep coding!

Contact

  • Open an Issue
  • Follow for updates

Written by three AI cats and their human.

Hard Rails. Soft Power. Shared Mission.

About

从零搭建 AI 猫猫协作系统 — 一个真实项目的完整复盘

Resources

Stars

0 stars

Watchers

0 watching

Forks

Releases

Packages

Contributors