cindy
worthwhileA multi-harness, multi-model desktop/mobile agent orchestrator that bundles Claude Code and Codex with shared memory and skills. Useful polish, not a paradigm
What it is
Cindy is an open-source desktop (Electron) and mobile (Expo/React Native) application that wraps existing AI agent harnesses — currently Claude Code and Codex — into a single interface with shared memory, skills, automation scheduling, and MCP integration. It runs locally, supports multiple models, and allows mid-task switching between harnesses.
How it differs from vanilla Claude
Claude Code alone is a capable CLI agent that can read files, execute commands, and edit code. Cindy adds a GUI, persistent memory across sessions, reusable skills, parallel multi-harness execution (e.g., Codex plans while Claude Code codes), and a mobile client. The base agent doesn't natively orchestrate multiple different agent harnesses or persist learned corrections across distinct agents.
Skill, plugin, or workflow shift?
Standalone-app — it is a desktop and mobile application with its own interface, login, and workflow. It orchestrates external agents, not a skill/plugin you add to an existing agent.
Devil's advocate — is this just complexity?
Claude Code (the base agent) already handles the vast majority of coding tasks with a single, coherent context. Adding a second harness and a UI overlay solves a problem most developers don't have: needing to switch between different agent implementations mid-task. The shared memory is a thin SQLite-like persistence over what Claude's own edit-context already does. Cindy bundles two command-line tools into an Electron shell with extra knobs — exactly the kind of complexity-for-its-own-sake wrapper that adds failure modes (submodule sync, platform tool binaries, pnpm version pinning) for marginal gain. If you only use Claude Code, Cindy adds zero new capability. The 'multi-model' claim is also hollow when local mode disables server capabilities and the hosted service costs money.
What would make it better
Drop the dependency on third-party harnesses entirely — build a native agent from scratch that justifies the orchestrator. Alternatively, ship a zero-install single-binary with no Node/pnpm prereqs. Until one of those happens, it's a leashed wrapper.
The honest case for it
If you regularly switch between Claude Code and Codex depending on context (e.g., planning vs. coding), shared memory and skills eliminate manual context transfer. The mobile client lets you check progress or trigger automations from a phone. For teams exploring multi-agent orchestration, Cindy provides a tangible playground that is open and forkable.
Who it's for
Audience fit
Depth and leverage for a technical engineer who wants to understand it and level up their workflow — not just offload work.
Value for someone who wants a more capable tool without the technical depth — accessible, does-it-for-you.
Engineers gain multi-harness orchestration, shared memory/skills across models, and a client you can fork. Vibe coders get a pretty launcher that bundles Claude Code and Codex, but the real leverage requires understanding harness interactions.