Repolode
Submit
KKKKhazix

KKKKhazix/agent-autoloop

A lightweight, zero-dependency scheduler that lets command-line AI agents like Claude Code and Codex work in the background and ask you for decisions via cards only when needed.

View on GitHub
Stars
3,938
Forks
715
Open issues
8
Last commit
7 hours ago

What is it?

Agent Autoloop turns command-line agents such as Claude Code and Codex from tools you must watch into background workers. A zero-dependency kernel handles scheduling, while event sources, agents, notification channels and delivery pipelines are plugins that can be swapped. Events from cron schedules, GitHub webhooks or HTTP calls are routed by rules in a local rules.json file to a chosen agent and working directory.

When an agent hits a decision it cannot make alone, such as deleting data, restarting a service or sending an external message, the system sends a card to Feishu, DingTalk, WeCom, Slack, Discord or a desktop notification. While waiting for a reply the process exits, so no agent process runs or consumes tokens. After you answer, the agent resumes the original session and finishes the work. It runs on macOS and Linux without a cloud server or public ports.

Agents integrate through a simple convention: they print a final result and optionally append a structured block to ask questions or schedule follow-ups. Any command-line tool that takes a prompt and prints a reply can be connected with a single config entry.

Key features

  • 01Rules in rules.json route cron, GitHub or HTTP events to a chosen agent and directory.
  • 02Waiting for human input consumes no process or tokens; sessions resume after a reply.
  • 03Decision cards reach Feishu, DingTalk, WeCom, Slack, Discord or desktop notifications.
  • 04Built-in plugins for Claude Code, Codex and ACP, plus a generic command adapter for other CLI agents.
  • 05A delivery pipeline reviews, waits for CI, merges and verifies a PR after the deliver label is added.
  • 06An append-only event log lets you trace why any run happened with a single command.

Use cases

  • Running a daily morning briefing from local notes and git commits.
  • Triaging new GitHub issues in a local checkout and notifying you for review.
  • Investigating production alerts and asking for approval before restarting services.
  • Monitoring a website and sending one alert on failure instead of repeated notifications.

Quick start

  1. 1

    Run the official install script from the README (requires Node 24, Git, curl and a logged-in agent CLI)

  2. 2

    Run a real test round with your local agent

    agent-autoloop try
  3. 3

    Check progress and usage after answering the card

    agent-autoloop status

Things to know

  • Supports only macOS and Linux.
  • Requires Node 24 and a pre-installed, logged-in command-line agent such as Claude Code or Codex.
  • The README advises using API keys rather than personal subscriptions for high-frequency automation.

In collections

Claude Code tools, skills & pluginsOpen-source AI agent frameworks & appsAI coding assistants & agentsMCP serversSelf-hosted appsCLI & terminal tools

Topics

#ai-agents#automation#claude-code#codex#cli#workflow#self-hosted#event-driven#mcp#nodejs

FAQ

What is agent-autoloop?
A lightweight, zero-dependency scheduler that lets command-line AI agents like Claude Code and Codex work in the background and ask you for decisions via cards only when needed. Agent Autoloop turns command-line agents such as Claude Code and Codex from tools you must watch into background workers. A zero-dependency kernel handles scheduling, while event sources, agents, notification channels and delivery pipelines are plugins that can be swapped. Events from cron schedules, GitHub webhooks or HTTP calls are routed by rules in a local rules.json file to a chosen agent and working directory.
Is agent-autoloop free to use? What license does it have?
agent-autoloop is open source under the MIT license. Check the license terms before using it in commercial products.
Is agent-autoloop still maintained?
Actively maintained. The last commit to agent-autoloop was on Nov 7, 2026.
What can I use agent-autoloop for?
Running a daily morning briefing from local notes and git commits. Triaging new GitHub issues in a local checkout and notifying you for review. Investigating production alerts and asking for approval before restarting services. Monitoring a website and sending one alert on failure instead of repeated notifications.
How do I install agent-autoloop?
Run the official install script from the README (requires Node 24, Git, curl and a logged-in agent CLI) Run a real test round with your local agent: agent-autoloop try Check progress and usage after answering the card: agent-autoloop status
Where can I get agent-autoloop?
The source code and installation instructions are on GitHub: https://github.com/KKKKhazix/agent-autoloop

Summary generated by AI from the project's README. GitHub stats refresh daily.

Related projects