Cindy treats Claude Code and Codex as swappable engines
Cindy is on GitHub Trending today at about 470 stars a day, now past 4,000 total. Its pitch is not a better coding agent. It is a layer that sits above coding agents and treats them as parts.
The README says it directly: "Models and harnesses mix freely and can switch mid-task while your workspace, memory, skills and tools stay continuous." Claude Code and Codex are the first two supported harnesses, and a native one is in the works. One task can be planned on one harness and model, run in parallel on another, and reviewed by a third. Memory and skills are shared across all of them, so a correction you make once in a Claude Code session still applies when the task moves to Codex.
The billing design is the smart part. You can authorize the Claude Code or Codex coding plan you already pay for and keep using it inside Cindy, so there is no second bill. You can also bring your own API keys, use local models, or pay for Cindy's own service. The desktop and mobile client is Apache-2.0. The backend service is separate and closed, which is worth knowing before you call it fully open source.
Why it matters: for a year the fight has been over whose harness you live in. Cindy bets the opposite way, that harnesses become interchangeable engines and the thing you actually keep is the memory, the skills and the workspace sitting on top. If that's right, the sticky layer moves up one level, and the labs' coding agents start to look like interchangeable backends. Cross-harness review is also a cheap version of evaluator independence: let a different model on a different harness check the work.
Repo: https://github.com/makecindy/cindy Site: https://cindy.app
← Back to all articles
The README says it directly: "Models and harnesses mix freely and can switch mid-task while your workspace, memory, skills and tools stay continuous." Claude Code and Codex are the first two supported harnesses, and a native one is in the works. One task can be planned on one harness and model, run in parallel on another, and reviewed by a third. Memory and skills are shared across all of them, so a correction you make once in a Claude Code session still applies when the task moves to Codex.
The billing design is the smart part. You can authorize the Claude Code or Codex coding plan you already pay for and keep using it inside Cindy, so there is no second bill. You can also bring your own API keys, use local models, or pay for Cindy's own service. The desktop and mobile client is Apache-2.0. The backend service is separate and closed, which is worth knowing before you call it fully open source.
Why it matters: for a year the fight has been over whose harness you live in. Cindy bets the opposite way, that harnesses become interchangeable engines and the thing you actually keep is the memory, the skills and the workspace sitting on top. If that's right, the sticky layer moves up one level, and the labs' coding agents start to look like interchangeable backends. Cross-harness review is also a cheap version of evaluator independence: let a different model on a different harness check the work.
Repo: https://github.com/makecindy/cindy Site: https://cindy.app
Comments