kodo

kodo coordinates coding agents for developers who have a clear goal and want work to continue unattended for hours. It divides implementation and review across agents and multiple context windows.

• Resumable work: Runs span multiple work cycles, with automatic checkpoints and logs so interrupted tasks can resume.
• Agent teams: An orchestrator delegates work to installed backends such as Claude Code, Cursor and OpenAI Codex, with separate architect and tester roles available for review.
• Workflow testing: A testing mode exercises user-facing features and probes edge cases, including preparing fixtures and testing tools.
• Code review: A separate review mode examines unnecessary abstractions, duplicated logic, dead code, usability and architecture.
• Specialist reviewers: Custom teams can include agents focused on interface design, security or performance.

kodo runs locally on macOS, Linux and Windows and is open source under the MIT license. It requires at least one installed agent backend. The orchestrator can use an installed CLI or the Gemini or Claude API; API orchestration sends requests to an external service, and workers follow their backend's account and provider arrangements.

Completion doesn't enforce independent approval. The default completion path asks the orchestrator to consult testers, but the orchestrator decides when to request and accept review. Agents run with full permissions and can access files beyond the project directory.