AI agents that ship code, end to end

Cloche turns LLM agents into a development pipeline. Point it at a task list, and it picks up work, runs agents in isolated containers, validates the output, and hands back clean git branches ready for review. No babysitting. No manual cleanup.

A DSL for your development workflow

A simple DSL describes agent steps, test suites, review gates, and retry logic in one readable file. Wire steps together by their outcomes and Cloche handles execution, parallelism, and failure routing.

Agents in containers, codebase stays safe

Every agent step runs inside a Docker container with a fresh copy of your project. Agents write code and iterate freely without touching your host filesystem. Results come back as git branches ready for review.

Agents, scripts, and human review

Not everything should be an LLM call. Chain an agent step into make test, gate merges on human approval, or fan out to parallel validation. The right tool for each stage.

Orchestration that runs itself

The orchestration loop continuously picks up tasks, runs workflows, and manages the full lifecycle. Read from GitHub Issues, Linear, or a local file — Cloche handles the rest. Multiple workflows run concurrently, each in its own container, each producing isolated results.

Any agent, one interface

Claude Code, Codex, or anything that reads a prompt and reports a result. Swap agents per step, set up fallback chains, or bring your own.

Ready to try it? Install Cloche and run your first workflow in minutes.