Clowder AI is an open-source platform designed to turn separate AI agents into a coordinated team. Instead of using Claude, GPT, Gemini, or other AI tools independently, Clowder provides a shared layer for communication, memory, task handoffs, review, and persistent agent identities.
The project is built around the idea that different AI models can take different roles. One agent can handle implementation, another can review the work, while others can focus on research or design. Clowder acts as the coordination layer rather than replacing the underlying AI agent tools.
Features
Clowder AI provides several components for coordinating multiple AI agents.
Multi-agent orchestration
Persistent agent identities
Shared memory
Cross-model review
Agent-to-agent communication
Structured task handoffs
Evidence and decision logs
MCP support
Plugin framework
Feishu and Telegram integrations
On-demand skills
Claude Code integration
Codex CLI integration
Gemini CLI integration
OpenCode integration
Desktop applications for Windows and macOS
Web interface
Self-hosted deployment
Optional Redis support
MIT license
The supported agent model is particularly interesting because Clowder does not attempt to replace the individual agent CLIs. Instead, it sits above them and provides the coordination layer.
Download Clowder AI v0.13.0 - Software Mirrors |
|---|
Clowder AI v0.13.0 for Windows |
Clowder AI v0.13.0 for macOS |
Clowder AI v0.13.0 Source Code |
Clowder AI v0.13.0 Release Notes:doc_kind: release-notes topics: [opensource, release, messaging, runtime, desktop] created: 2026-09-01 release: v0.13.0 Clowder AI v0.13.0
Highlights
fast lane now have an 18-case joint acceptance contract. Stop/Steer ownership, proactive final output, same-thread fan-out, and scheduler wake handling are more reliable.
context window on resume; OpenCode model aliases, output limits, and silent completions are recovered correctly; Kimi >= 0.30 resume compatibility and Claude partial-output lifecycle are fixed.
cursor repair no longer revives stale badges, streaming keeps the viewport anchored, and Context Window / Session State settings are simpler. Uninitialized projects now show neutral onboarding instead of a false runtime error.
idempotent, Windows startup fails closed when Redis is unavailable, Windows process groups are terminated correctly, and desktop update recovery now provides clearer diagnostics and retry paths. What's ChangedMessaging, plugins, and coordination
suite.
same-thread A2A fan-out.
authenticated binding. Providers, sessions, and runtime
and added safe manual session sealing.
post-tool no-output completion recovery.
bootstrap, CLI incompatibility diagnostics, and manual-cancel slot liveness. Hub and settings
labels for dead configuration.
creation when no accounts exist.
Reliability, Windows, and desktop
is unavailable.
desktop bundle.
exclusions with an audited 16 re-admit / 25 retain decision. Compatibility and Upgrade Notes
and configuration remain in place. Node (
Redis or explicitly choose the volatile
such as
called the internal hook routes with
entrypoints inject the required opt-in. Custom API launch commands that expect IM connectors to autostart must set
not an active notification path; PR review feedback uses GitHub API polling. Downloads
The desktop app can discover this release automatically. Manual downloads remain available from the GitHub release page. CommunityThank you to the contributors whose changes landed in this release, including @mindfn, @ldn9596, @solcao-git, @malt77, @TERRYYYC, @masterkunm, @QwQ02009, @whutzefengxie-ops, and @funkdog. Full changelog: v0.12.0...v0.13.0 Clowder AI v0.13.0
重点更新
联合验收场景约束。Stop / Steer 球权、主动
OpenCode 的模型别名、输出上限和静默完成得到修复；Kimi >= 0.30 resume 兼容、Claude partial output 生命周期也已补齐。
流式输出时视口能稳定吸底；Context Window / Session State 设置更清晰。未初始化项目 会显示中性引导，不再误报运行环境损坏。
不可用时会 fail closed；Windows 子进程组可被正确回收；桌面更新增加了更清楚的诊断 和重试路径。 完整变化消息、插件与协作
Provider、Session 与 Runtime
手动 session seal。
诊断和手动取消后的 slot 存活。 Hub 与设置
可靠性、Windows 与桌面端
“16 项重新纳入 / 25 项继续保留”的审计。 兼容性与升级说明
(
明确接受易失存储时才主动使用
opt-in；自行直接启动 API、且希望 IM connector 自动启动时，请显式设置
通道；PR review 反馈走 GitHub API polling。 下载
桌面端可以自动发现本版本，也可以从 GitHub release 页面手动下载。 社区贡献感谢本版本所有贡献者，包括 @mindfn、@ldn9596、@solcao-git、@malt77、 @TERRYYYC、@masterkunm、@QwQ02009、@whutzefengxie-ops 和 @funkdog。 完整变更：v0.12.0...v0.13.0 |
Performance and Compatibility
Clowder AI is available as a desktop application for Windows and macOS. The project also supports running from source, making it possible to use the platform without relying on the packaged desktop applications.
The current source setup requires Node.js 20 or newer and pnpm 9 or newer. Redis 7 or newer is optional, with a memory mode available when Redis is not used.
The project is under active development. The v0.13.0 release focused on stability improvements around agent messaging, provider and session recovery, the Hub interface, and desktop functionality. The release also provides Windows installers and portable packages plus Intel and Apple Silicon macOS builds.
Clowder is more demanding conceptually than a normal AI chat application. Its main value depends on actually using multiple agents and allowing them to communicate and share work. Users who primarily work with one AI model may not need the additional orchestration layer.
System Requirements
For running Clowder AI from source, the project lists:
Node.js 20 or newer
pnpm 9 or newer
Git
Redis 7 or newer, optional
Internet access for connecting external AI providers and services
Prebuilt desktop releases are available for Windows and macOS. The current release provides Windows x64 installers and portable packages, along with Intel and Apple Silicon macOS builds.
Pros and Cons
Pros
Open source under the MIT license
Supports multiple AI model families
Persistent agent identities
Shared memory between agents
Built-in cross-model review
Agent-to-agent communication
MCP support
Supports Claude Code, Codex CLI, Gemini CLI, and OpenCode
Self-hostable
Windows and macOS desktop applications
Portable Windows version available
Apple Silicon and Intel Mac support
Active development
Cons
More complex than a conventional AI assistant
Requires multiple AI tools to get the most value from its architecture
Source installation requires Node.js and pnpm
Some deployments may require additional infrastructure such as Redis
The project is still relatively young and evolving quickly
Users need to understand agent permissions, provider authentication, and coordination workflows
How to Install
Windows and macOS
Download the appropriate desktop release from the project's GitHub Releases page. The packaged application includes the required components, so a separate pnpm installation is not required.
From Source
Clone the repository and install the dependencies:
git clone https://github.com/zts212653/clowder-ai.git
cd clowder-ai
pnpm install
pnpm build
cp .env.example .env
pnpm startThe local interface is then available at:
http://localhost:3003After starting Clowder, provider and CLI accounts can be configured from Hub > System Settings > Account Configuration.
Final Verdict
Clowder AI takes a different approach from conventional AI assistants. Instead of trying to make one agent handle everything, it provides infrastructure for several AI agents to work together while retaining their own identities, memories, roles, and model capabilities.
Its strongest feature is the coordination layer. Shared memory, cross-model review, agent-to-agent messaging, MCP integration, and structured handoffs address problems that appear when using several AI coding and reasoning tools simultaneously.
The trade-off is complexity. Clowder makes more sense for users who regularly work with multiple AI agents than for someone who only needs a single chatbot or coding assistant. The project is also evolving quickly, so workflows and capabilities can change between releases.