Twindem adds human approval and records to AI coding work
Twindem is an app for using command-line like Claude Code and Codex inside a controlled development flow. Work moves through idea, plan, , review, , and done, with a human approval gate before each next phase. One AI tool writes the code and another AI tool reviews it, while the review and fix cycle is limited so it cannot continue endlessly.
GitHub Projects or Jira acts as the main work board instead of relying on the app’s own private state. Each step leaves evidence, so it is possible to look back and see what the AI changed and who approved it. The design is local-first: the AI tools run on the user’s computer as , secrets are stored in the macOS keychain, and there is no account, cloud upload, or telemetry.
The project says the user’s repository stays on their machine. It is still before version 1.0, and it is meant to support human review rather than replace it.
Key points
- Twindem works with command-line such as Claude Code and Codex.
- Each phase of the work needs human approval before it moves forward.
- One AI tool implements the change and another reviews it.
- GitHub Projects or Jira is treated as the main source of work status.
- The app is local-first, with no account, cloud upload, or telemetry according to the project.