ai-coworkers

ai-coworkers is a daemon with a job description rather than a chat interface. It is a long-running process that holds a job, wakes on a clock or webhook, reads its sensors, and acts inside a role written in markdown with its own responsibilities and boundaries. When it needs a human, it asks and keeps working. The project ships with seven roles: triage engineer, PR reviewer, project manager, scribe, incident RCA, changelog, and monitoring. A coworker is configured as a directory of markdown files such as ROLE.md, RESPONSIBILITIES.md, AUTHORITY.md, BOUNDARIES.md, RITUALS.md, and RELATIONSHIPS.md, with no YAML or code changes required to tune behavior. Every action is checked against BOUNDARIES.md before it executes, and writes are dry-run until the operator passes --live. A single ask tool routes questions to the operator, a peer coworker, Slack, or a GitHub PR, and questions persist in state/questions.md until answered. A tick that finds nothing new returns before any model is called, including triage. Memory is stored in six tiers in local SQLite with FTS5, and the harness can run MCP servers, Hermes/OpenClaw skills, Vercel Eve agent folders, and ACP coding agents. The codebase is small enough to read cover-to-cover at about 7.5k lines of TypeScript and is MIT-licensed.