Agent Autoloop is an open-source TypeScript project for running command-line coding agents in the background. A local scheduler dispatches tasks to tools such as Claude Code and Codex, then sends an approval request through Feishu, Slack, Discord, DingTalk, WeCom, or desktop notifications when a task requires human input.
The repository currently has 4 stars. Its architecture separates a small scheduling kernel from replaceable plugins for event sources, agent adapters, notification channels, and CI/CD delivery.
Why terminal agents stall
Interactive coding agents usually depend on an open terminal and a person available to answer questions. Closing the session or leaving a prompt unanswered can interrupt the task.
Agent Autoloop turns each task into a bounded invocation called a round. The agent starts, processes the available work, records its final reply, and exits. A later event or human response can resume the original conversation through the relevant adapter.
- Local deployment: The project supports macOS and Linux without requiring a hosted control plane or public port.
- Command adapters: Any CLI tool that accepts a prompt and returns a final response can be configured through a command wrapper.
- CLI operations: There is no web dashboard. A status command reports progress.
- Existing credentials: Agents use the authentication already configured in their respective CLI tools.
From trigger to resumed session
- An event such as a cron trigger, GitHub issue, or production alert enters the scheduler in a normalized format.