Breakwater

Breakwater is a local service that acts as an automated on-call assistant for software engineering teams. The name comes from the idea of a breakwater barrier: by using a large language model and historical incident data, it intercepts most incoming issues automatically so that human engineers only have to step in for decisions that genuinely require judgment.

The project connects to three platforms: Jira for ticket management, GitHub for repository issues, and Lark (a team messaging app popular in China). When a new issue arrives in Jira or GitHub, Breakwater can record it, run an automated analysis using an AI coding agent called Codex, search past Jira tickets for similar problems and their solutions, and post a summary comment. In Lark group chats, it can link discussions directly to the relevant issue so context is not lost across tools.

Breakwater also includes a development mode. If the AI agent identifies a clear coding task and the local repository and permissions are in place, it can attempt end-to-end code writing and submit the result upstream. This is an optional path controlled through configuration; it is not the default behavior.

Setup involves a YAML configuration file where you specify your Jira URL and access token, which GitHub repositories to monitor, which Jira projects to watch, queue concurrency, proxy settings, and optional admin credentials for a management web panel. The service runs locally on your machine, starts and stops via command-line commands, and exposes a status page at a local port. All prompts and internal skills that guide the AI agent are plain text files in the project, so they can be edited without touching code.

The project is relatively new, has few stars, and its documentation is written primarily in Chinese. It appears designed for teams that already use Codex as a coding agent and want to wire it into their Jira and GitHub issue workflows to reduce manual triage work.