An Open-Source AI Agent That Mixes Claude Code and Codex Mid-Task
You've probably got a favorite coding agent by now—maybe Claude Code, maybe Codex—and you've probably noticed it has blind spots. It's good at some things and stubborn about others, and you can't just swap in a different model when you hit a wall. Cindy is an open-source agent that takes a different approach: it lets you mix harnesses and models freely, switching between them mid-task while your workspace, memory, skills, and tools stay continuous.
What It Does
Cindy is an AI agent that runs locally on your own machine, using your real files and logged-in apps. The first harnesses it supports are Claude Code and Codex, with more on the way and a native harness in development. The interesting part is that models and harnesses mix freely—one task can be planned, executed in parallel, and reviewed by agents on different harness × model combinations, and everything switches mid-task without losing context.
The repository here is the open-source client: an Electron desktop app, an Expo/React Native mobile client, and shared packages covering auth, device-link, agent orchestration, and model providers, organized as a pnpm monorepo. The backend service lives in a separate repository and isn't part of this monorepo. The client runs on Node.js 22.x and pnpm 10, and it's licensed under Apache-2.0.
According to the README, she can also drive your browser, computer, and phone, and take work from IM and schedules—so it's not just a coding assistant but something closer to a general-purpose agent that does real work on your computer.
Why It's Cool
The mid-task switching is the real hook. Most agent setups lock you into one harness for a session. Cindy treats harnesses and models as interchangeable parts of a single continuous workspace, which means you can let one model plan and another execute or review. That's a genuinely different orchestration model, and it's the kind of thing that only works if memory and tools are decoupled from any single harness—which is exactly what the README describes.
You bring your own billing. Sign in to the official Cindy service, authorize the Claude Code or Codex Coding Plan you already pay for (no duplicate bill), connect your own API keys, or run local models. That flexibility matters if you've already got a subscription you don't want to waste.
Open source here means changeable, not just visible. The README makes the distinction explicitly, and the list backs it up: correct her once and she remembers across harnesses, teach a skill once and reuse it everywhere, schedule recurring work that reports back, wire in internal tools via MCP. Plugins for reshaping features and UI are listed as in the making, as is sharing skills with your team—so treat those as roadmap, not shipping features.
Sensible packaging. The tool binaries (claude-code, codex, ripgrep) aren't committed to the repo—they're downloaded per platform during pnpm install, and the Android platform-tools binaries are fetched with a pinned version and sha256 verification before Windows packaging. That's a tidy way to keep the monorepo clean without shipping stale binaries.
How to Try It
The easiest path is to grab a build from the official download page rather than building from source:
- Head to cindy.app/download and pick the build for your platform.
- If you're on Linux, the README points to a dedicated guide covering Ubuntu, Arch Linux, and Omarchy—installation, updates, keyring setup, and migrating from a manual install.
- Sign in to the official service, authorize your existing Claude Code or Codex plan, or plug in your own API keys.
If you'd rather build it yourself, clone the repo and use the pnpm workspace:
git clone https://github.com/makecindy/cindy
cd cindy
pnpm install
Note that pnpm install pulls down the platform-specific tool binaries, so expect it to fetch a few things. From there, apps/desktop and apps/mobile are the two clients to look at. The README also points contributors to CONTRIBUTING.en.md—worth reading before you start poking around the monorepo.
Final Thoughts
Cindy is aimed at developers who already use coding agents and want more control over which one does what—and who'd rather keep their existing subscriptions and local models than start over with a new billing relationship. The mid-task harness switching is the standout idea, and the local-first, open-source client makes it something you can actually audit and modify. Some of the most interesting pieces (plugins, skill sharing, the native harness) are still in the making, so go in expecting a project that's ready to use today but clearly still growing. If that sounds like your kind of tool, the source is yours to shape.
Follow @githubprojects for more developer tools and open source projects.