buagi/ace

ACE, short for Agentic Coding Environment, is a command line tool that installs a crew of twelve AI agents meant to build software on your behalf with very little supervision. You write down your goals in a file, run one command, and the crew plans the work, writes the code, tests it, reviews it, merges it, and deploys it on its own. One agent acts as the orchestrator, breaking your goals into small tasks and handing them out, but never writing code itself. Another agent, the implementer, actually builds each task. A verifier agent runs the project's tests and checks that the code matches what was claimed. Depending on how risky a change looks, either a single reviewer approves it or a full panel of four separate reviewers checks it from different angles, covering correctness, user experience, coding standards, and whether the change fits the project's overall direction. A change only gets committed once the verifier passes and the relevant reviewers approve, and the tool itself merges the result rather than letting an agent merge its own work. To avoid making things up about your codebase, the system connects to a few external tools that ground it in real information: one that maps how code pieces connect to each other, one that tracks live symbol definitions, and one that pulls current library documentation. There is also a fully automatic mode that keeps a running list of goals and tasks, builds the next item, fixes failing tests by reading the actual error logs, merges when everything passes, and deploys before picking up the next task. The crew of specialist agents runs on DeepSeek, a lower cost AI model, while the main overseer agent defaults to Claude Opus but can be swapped out, and the whole system can run using only a DeepSeek API key if you prefer to avoid a Claude subscription. It is built to run on Linux systems such as Fedora Silverblue or Arch, without needing root access.