Cindy
An open-source desktop and mobile agent for completing work on your computer and signed-in apps.
- Source repo
- makecindy/cindy
- Stars
- ★ 9.3k
- Last updated
- today
- License
- Apache-2.0
- Primary language
- TypeScript
- FA score
- 52/100 · Major gaps
At a glance
- How it runs
- Works with
- Portable with changesCodex · Claude Code
- Cost
- Free tier plus a paid hosted plan
- Setup effort
- Medium · a few setup steps
- You'll need
- Typical use
- A developer with an existing Claude Code or Codex Coding Plan who wants to use that authorization in one client and switch harnesses during a task.
- Not a fit if
- Teams needing a fully self-hosted backend; this repo is client-only
- Users who need a fully offline setup without vendor cloud services
- Teams unwilling to use a Cindy cloud account or official default endpoints
- Source review
- 52/100 · Major gaps
What does this agent do, and when should you use it?
Cindy is an open-source AI agent client delivered as an Electron desktop app, an Expo/React Native mobile app, and shared packages in a pnpm monorepo. It runs on the user’s own machine and is described as working with real files and logged-in applications. Its first documented harnesses are Claude Code and Codex, with task-time switching across harnesses and models while workspace, memory, skills, and tools remain continuous. The repository also describes browser, computer, and phone control, plus work intake from IM and schedules; the backend service is explicitly outside this repository. Users can sign in to the official Cindy service or choose Skip Sign-In for local agents, with server-backed capabilities unavailable in that state.
The Cindy client brings Claude Code, Codex, models, and tools into one task environment for work involving local files and signed-in apps. A task can be planned, executed in parallel, and reviewed by agents using different harness × model combinations while its workspace, memory, skills, and tools continue across those stages. apps/desktop contains the desktop client, apps/mobile contains the mobile client, and packages/* holds shared capabilities including auth, device-link, agent orchestration, and model providers. It also documents MCP connections to internal tools and business systems, along with Automation for scheduling recurring work and reporting back.
- A developer with an existing Claude Code or Codex Coding Plan who wants to use that authorization in one client and switch harnesses during a task.
- An individual or team member who needs an agent to work with project files and applications already signed in on their computer.
- An organization connecting internal business systems and tools to task workflows through MCP.
- An operations user scheduling recurring work to run automatically and return a report.
- A user who needs both Cindy’s desktop client and its mobile client.
How do you install or deploy this agent?
Install Node.js 22.x, pnpm 10.x, and Git LFS, then run:
git clone https://github.com/makecindy/cindy.git
cd cindy
git submodule update --init --recursive cindy-protocol
git lfs pull
pnpm installThe repository states that a public checkout needs the cindy-protocol submodule. Remote development against the official service requires your own Cindy cloud account and existing login state.
How do you use this agent?
For a Mainland China Cindy account, run: pnpm restart:desktop:remote --region=cn. For a global account, run: pnpm restart:desktop:remote --region=global. After signing in, you can continue existing sessions; alternatively choose “Skip Sign-In” on the login screen to run local agents. That state is shown as “Not signed in” and does not include server-backed capabilities. The supplied documentation does not provide a public HTTP API, SDK invocation example, or local server startup command.
What are this agent's strengths and limitations?
- It documents Claude Code and Codex as first-class harnesses in one task environment, including switching harnesses and models mid-task.
- Electron desktop and Expo/React Native mobile clients share capabilities such as auth, device-link, agent orchestration, and model providers.
- It supports several documented model-access routes: the official service, existing Claude Code/Codex Coding Plan authorization, user API keys, and local models.
- It explicitly offers a local-agent Skip Sign-In path and MCP connectivity for internal tools and business systems.
- This repository is client-only; the backend service is in a separate repository, so it does not document a complete self-hosted server deployment.
- Skip Sign-In disables server-backed capabilities, reducing the available feature set.
- Building from source requires Node.js 22.x, pnpm 10.x, Git LFS, and the cindy-protocol submodule; pnpm v11 is not yet supported.
- Default endpoints target official Cindy cloud services and the official CDN, while remote development requires a Cindy cloud account and existing login state.
- The open plugin marketplace and team sharing for skills are still described as in the making.
How does this agent compare with similar options?
Compared with using Claude Code or Codex separately, Cindy documents them as harnesses that can switch within one task while preserving the workspace, memory, skills, and tools. Whether it uses the official service, user API keys, or local models depends on the selected setup.
Key facts side by side with the most closely related agents.
| Agent | Source review | Form / cost | Stars | Updated | Language | Full support on |
|---|---|---|---|---|---|---|
| Cindy This agent | 52 · Major gaps | Desktop appFreemium | ★ 9.3k | today | TypeScript | Codex · Claude Code |
| MyAgents | 57 · Major gaps | Desktop appFree + model costs | ★ 591 | 5d ago | TypeScript | Codex · Claude Code · Claude API |
| HermesOffice | 57 · Major gaps | Desktop appFree + model costs | ★ 624 | 3d ago | TypeScript | OpenAI API · Claude API |
| Ankita Desktop AI Workspace | 54 · Major gaps | Desktop appFree + model costs | ★ 14 | 5d ago | JavaScript | ChatGPT · OpenAI API |
How does FollowAgents rate this agent?
Why each dimension lost points
Evidence: README explicitly states default connection to official cloud services and discloses telemetry (TapDB) and heartbeat data, but lacks user confirmation mechanisms or data flow visualization. Dependency security: package.json has overrides and patchedDependencies, CI verifies submodule source, but no dependency vulnerability scanning evidence. External effects: README mentions driving browser, computer, and phone, but no permission controls. Rollback: not mentioned. Source attribution: DCO and contribution guidelines present, but publisher unverified. Deductions: lack of user confirmation, least privilege, and rollback implementation evidence.
Evidence: CI configures multi-platform tests (Linux/Windows) and sharding, but no test results or coverage. Dependency availability: pnpm lockfile and overrides present, but no availability guarantees. Failure messages: no specific error handling examples. Deductions: static review cannot verify test execution; failure message evidence insufficient.
Evidence: README describes multiple use cases (desktop, mobile, browser, IM) and capability boundaries (local running, supports Claude Code and Codex). Trigger precision: no specific trigger mechanisms. Environment fit: multi-platform and Node version requirements. Deductions: trigger precision evidence insufficient.
Evidence: README provides clear architecture and documentation index, installation steps, Apache-2.0 license, contribution guidelines. Naming stability: no version history. Examples and FAQ: basic examples but no FAQ. Known limitations: some features in progress. Versioning changelog: absent. Maintenance responsibility: SECURITY.md and contribution guidelines. Deductions: missing changelog and FAQ.
Evidence: README describes output capabilities (completing tasks, reporting), but no specific output examples. Marginal value: emphasizes open-source and customization, but no comparison with other tools. Cost-benefit: mentions free use and bring-your-own models, but no detailed cost analysis. Deductions: output usability evidence insufficient.
Evidence: README claims partially supported by docs (architecture, security), but no independent verification. Cross-source corroboration: no external references. Fact-inference separation: README distinguishes facts and plans, but not explicitly labeled. Deductions: lack of independent verification and external corroboration.
- Default connection to official cloud services may raise data privacy and network dependency concerns.
- Telemetry data, though not containing chat content, is associated with account ID; users should be informed.
- Dependencies are downloaded from third parties; despite SHA256 verification, supply chain security should be ensured.
Recent releases
Latest entries from the project's GitHub Releases. Notes are quoted from upstream as written, not rewritten or tested by FollowAgents.
-
Release notes excerpt
Desktop release 0.1.126 更新说明 🖥️ 「迁移助理」把旧电脑搬到新电脑 新电脑可从同账号另一台电脑一次性迁移插件与技能、常用设置、供应商和 SSH 主机;需在本机完成的登录与权限会整理成清单逐项引导。密钥要旧电脑当场允许后加密传输,登录与私钥不会复制。 🔁 从开工到合并,/pr 一路跟进 GitHub 插件新增 /pr 技能:自动开工作区、推送前检查并跟到合并,只在有新失败、新留言或可合并时提醒,也能用指定电脑和模型定时跟进。PR 卡片像标题栏一样显示检查、评审和合并状态,手机同样可用。 @dashhuang 🧩 插件卡片更丰富,插件也能照看伙伴 插件卡片新增模型下拉、应用图标、整列宽度与实时状态位,手机上的卡片与电脑端一致;GitHub 链接可直接打开。经授权的插件还能查看主人伙伴的状态、代为回答伙伴等待的问题,并读取最近的对话;插件生成的任务标题也只用用户原话。 @GaoWeiLiuXD @dashhuang @xd-bobo 💬 伙伴群聊更清楚、更稳 群聊会说明当前不可用的工具和附件,能力提示不再重复打扰;结算被拒时已生成的回复不会丢失。委派卡片状态统一,已读名单会显示领取执行的伙伴,轮流发言可逐项跳过,引用跳转也不再带动输入框。 🖼️ 壁纸下的界面更通透了 …View on GitHub ↗ -
Release notes excerpt
Desktop release 0.1.125 更新说明 📱 手机上看清并操作设备画面 手机远程桌面现在能查看并操作插件提供的设备画面(如模拟器):悬浮玻璃按钮汇集设备按键、方向、缩放和独立设置,可跟随手机转动或手动旋转;方向、触摸和掉线状态也已修正。 @dashhuang 🧩 插件能力与授权 账号头像对所有插件开放,头像下载走更安全的通道;插件能力改为按声明与用户授权统一使用,本地开发和 Forge 安装也更顺畅;插件调用会归属当前任务,卡片在对应任务里打开。 @GaoWeiLiuXD @dashhuang 💬 伙伴群聊更好用 伙伴群消息新增已读双勾和已读名单;折叠屏与宽屏下群聊保留左侧任务列表;群邀请链接改用分享服务,侧栏头像恢复,跳到底部和离底状态也更稳定。 @GaoWeiLiuXD @nanaco666 @zhaodengfeng @zqchris 💻 伙伴与远程 Agent 更稳 远程 Agent 现在使用任务所在电脑的 Cindy 技能;远程电脑掉线时会自动切到伙伴的下一个候选模型,而不是直接报错;伙伴的成果文件也与回复内容对齐了。 @dashhuang 🔌 模型与供应商 …View on GitHub ↗ -
Release notes excerpt
Desktop release 0.1.124 更新说明 💻 多台电脑与伙伴跨电脑 Agent 现在能看到账号下所有电脑和 SSH 主机的实时资源(CPU、内存、磁盘、在跑的任务);伙伴的技能文件会随换电脑保留,也能搜到其他电脑上的伙伴对话。 @elfcv @zqchris 🙋 代输入更顺手 代输入窗口最小化后会自动还原;主人可以直接在网页上登录,填好后卡片随之收起;浏览器页面读不到时,会把主人带到窗口前继续操作。 @dashhuang ⌨️ 手机远程键盘新布局 折叠屏手机上的远程桌面键盘重新设计,默认使用全键盘与触控板,剪贴板和快捷键集中到顶行;复古皮肤也换成新的键位布局,按键状态更准确。 @dashhuang 💬 手机群聊与分享更清楚 群聊的运行状态按话题分开显示,引用回复与私聊保持一致,并新增群消息通知设置;从系统分享进来的内容不再打断当前页面,侧栏也能区分群聊。 @GaoWeiLiuXD @zqchris 🔌 供应商、模型与用量 可以按自定义端点名称搜索供应商,模型回退后能重新选回原来的模型,共享的供应商组也能正常使用;订阅余量小组件支持置顶和隐藏账号。 @DavidShenXD @MagicLizi @WZBbiao @dashhuang @xuzini-xzn 📎 附件与文档阅读 …View on GitHub ↗