Shep

Shep is an open source orchestrator that runs several AI coding agents in parallel and carries each feature from a one sentence request to a pull request. It works with Claude Code, Cursor, Gemini or any agent that has a command line interface, and it runs locally on Node 22 or later, opening a dashboard on localhost. Every feature gets its own git worktree, branch and agent session, so parallel jobs do not overwrite each other or disturb the developer's own checkout. An optional spec driven mode has the agent write requirements, research and a plan as YAML that the developer approves before any code is written. After implementation, Shep commits, pushes and opens a draft pull request, then watches the CI pipeline and has the agent read the logs and push fixes when a check fails. Merging stays with the developer unless a flag is set to allow it. It suits developers and small teams who already pay for one or more coding agent subscriptions and want to run many tasks at once from a single dashboard, with the output arriving as ordinary pull requests that go through the team's usual review process. Shep is free and MIT licensed. It does not resell model access, so users bring the agent subscriptions or API keys they already have. It is started with a single npx command.