Open Source Multi-Model AI Platform

AI Teams That
Grow With You

Not just agents — a team that learns, evolves, and grows alongside you. Plugins, self-awareness, self-evolution. All in the open.

Grow Together. Build Forever.

Quick Start View on GitHub
6
Agent CLIs
60+
Built-in Skills
3
Languages
MIT
License

Why Clowder?

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

You have Claude, GPT, Gemini — but using them together means you become the router: copy-pasting context, tracking conversations, losing hours to middle management.

Most frameworks help you call agents. Clowder helps them work together — with persistent identity, shared memory, cross-model review, and collaborative discipline.

Not a framework. Not a wrapper. A platform layer that turns isolated AI agents into a team.

What It Does

Click a capability to explore

Multi-Agent Orchestration

Route tasks to the right agent — Claude for architecture, GPT for review, Gemini for design — in one conversation. Each agent keeps its role, personality, and memory across sessions.

  • ✓ 6 Agent CLIs supported (Claude Code, Codex, Gemini CLI, Antigravity, opencode, Kimi)
  • ✓ @mention routing — messages go to the right agent automatically
  • ✓ Thread isolation — context stays clean across features
Multi-agent chat with rich blocks

Shared Memory & Evidence

Evidence store, lessons learned, decision logs — institutional knowledge that persists across sessions, context compressions, and model switches.

  • ✓ Graph-based evidence store with semantic search
  • ✓ Anti-compression identity — agents survive context resets
  • ✓ Lessons learned auto-extraction and retrieval
Mission Hub dashboard

Cross-Model Review

Claude writes code, GPT reviews it. Different cognitive styles catch different bugs. Built-in, not bolted on.

  • ✓ Automated SOP: quality gate → request review → merge gate
  • ✓ No self-review — cross-family review is enforced
  • ✓ Vision guardian checks against original requirements
operator Bootcamp onboarding

Skills Framework

On-demand prompt loading. Agents load specialized skills (TDD, debugging, review, security) only when needed — keeping context clean and responses focused.

  • ✓ 60+ built-in skills covering development lifecycle
  • ✓ Skills loaded on-demand, not stuffed into system prompt
  • ✓ Custom skill creation with manifest registration
Cat Leaderboard

Agent-to-Agent Communication

Async messaging with @mention routing, thread isolation, and structured handoff — agents coordinate directly, not through you.

  • ✓ Five-tuple handoff protocol: What / Why / Tradeoff / Open Questions / Next Action
  • ✓ Cross-thread messaging with coordination lifecycle
  • ✓ Ball-passing discipline with hold/wake patterns
Signal Inbox

MCP Integration

Model Context Protocol for tool sharing across agents, including non-Claude models via callback bridge. Shared tools, shared context.

  • ✓ MCP servers shared across all agent CLIs
  • ✓ Callback bridge for non-MCP-native agents
  • ✓ Limb system for external service integration
Signal study area

More Than Coding

Your AI team can do a lot more than write code

Collaborative Coding

TDD workflows, cross-model review, automated SOP. From idea to shipped code with real governance.

Voice Companion

Hands-free mode with per-agent voice. Talk to your team while running, commuting, or just thinking out loud.

Game Night

Werewolf, Pixel Cat Brawl — real games with your AI teammates. Stress-tests the same A2A messaging that powers work.

Signals Research

Curated AI research feed with multi-cat analysis. Your team reads papers and generates podcast-style discussions.

Multi-Platform

Chat from Feishu (Lark), Telegram (coming soon), or GitHub PR comments. Each cat replies as a distinct card.

Self-Evolution

The team reflects on its own processes, learns from mistakes, and improves — without being told. Lessons compound.

Architecture

Three layers, each a multiplier

Model Layer

Reasoning, generation, understanding

Claude · GPT · Gemini · Any LLM
Agent CLI Layer

Tool use, file ops, commands

Claude Code · Codex · opencode · ...
Platform Layer

Identity, collaboration, discipline, audit

Clowder AI

Models set the ceiling. The platform sets the floor. Each layer is a multiplier, not addition.

Quick Start

Up and running in minutes

Terminal
git clone https://github.com/zts212653/clowder-ai.git cd clowder-ai pnpm install pnpm start

Requires Node.js 20+, pnpm 9+. Redis 7+ optional (use --memory to skip).

⚡ Slow? Use pnpm start --quick 🔌 No pnpm? bash <(curl -fsSL .../install.sh) --start
Download for Windows

Loading latest version...

Run the .exe installer, then launch Clowder AI from Start menu.

Apple Silicon Intel

Loading latest version...

Unsure? The Apple menu → About This Mac shows your chip. Open the .dmg, drag to Applications, and launch.

Full setup guide: Docs → Quick Start

Get Started in 3 Steps

Configure credentials, invite your team, and start collaborating.

1

Add Your Accounts

Open Settings and add your AI provider credentials. Each cat gets its own account.

2

Invite Your Cats

Add team members with their own model, identity, and personality. They'll share context and collaborate.

3

Say Hi

Open a thread and send your first message. Watch your AI team come alive and start working together.

Roadmap

Growing in the open. Real dates, real PRs.

Shipped In Progress Planned
Mar Apr May Jun Jul Aug Sep →
Platform
Infrastructure
Foundation
Skills & MCP
A2A Contract
Plugin Framework
Intelligence
Memory & AI
Observability
Memory v2
Perception
Shared Memory
Cold-Start
Experience
UI & Surface
Desktop App
Homepage
IM Connectors
Console v2
Telegram
Request a Feature →

Cats & U

Clowder AI is extracted from Cat Cafe — a production workspace where four AI cats collaborate daily on real software. Every feature has been battle-tested.

The name clowder is the English collective noun for a group of cats — like "a murder of crows" or "a pride of lions." It also hides a small easter egg: clowder sounds a lot like cloud.

"Every idea deserves a team of souls who take it seriously."