agent-autoloop

让你的命令行 Agent 在后台自己接活、自己跟进到底;平时不烦你,遇到拿不准的事才发卡片找你拍板。

What is agent-autoloop?

让你的命令行 Agent 在后台自己接活、自己跟进到底;平时不烦你,遇到拿不准的事才发卡片找你拍板。

Overview of agent-autoloop

What it is

Agent Autoloop is a lightweight automation middleware for command-line AI agents. It runs on a single computer (Windows, macOS, or Linux) with no cloud server or public port required. The tool lets agents work autonomously in the background, only interrupting the user when a decision is needed—via a card pushed to messaging platforms or desktop notifications. It consists of a zero-dependency kernel for task scheduling and a plugin system for event sources, agents, notification channels, and delivery pipelines. The project is open source under the MIT license.

Key features

  • Runs on a single machine with one command; no cloud server or public port needed.
  • Plugin architecture: event sources (GitHub Webhook, GitLab Webhook, HTTP, cron, issue polling), agents (Claude Code, Codex, ACP, DeepSeek Harness, ZCode, generic command), notification channels (desktop notification, Feishu, DingTalk, WeCom, Slack, Discord), and delivery pipelines are all swappable.
  • Agents only need to output the result in their last reply; no SDK or source code modification required.
  • When an agent encounters an uncertain action (e.g., deleting data, restarting a service, spending money), the kernel sends a card to the user and pauses the agent, consuming no tokens or processes while waiting.
  • Rules are written in a JSON file (rules.json) without glue code; changes take effect within seconds without restarting the middleware.
  • Supports session resumption: if a session is interrupted, the kernel retries the original session instead of starting a new one.

Use cases

  • Automated daily briefings: schedule a cron job to have an agent summarize notes and Git commits every morning.
  • GitHub issue triage: automatically check new issues, modify code locally, and notify the user for review.
  • Online alert investigation: receive error alerts, have the agent analyze the code, and ask the user before restarting services.
  • User feedback processing: re-read feedback, fix issues, verify, and prepare replies, only asking the user when a decision is needed.
  • Website monitoring: check site status every 5 minutes, hand off anomalies to an agent, and avoid duplicate notifications for persistent issues.
  • PR automated delivery: after a PR is labeled deliver, the agent reviews, waits for CI, merges, waits for release, and verifies the live version.

Getting started

Installation requires Node 24 or newer, Git, and a logged-in command-line agent. macOS/Linux also need curl. No GitHub account required.Windows PowerShell:powershellirm https://raw.githubusercontent.com/KKKKhazix/agent-autoloop/main/install.ps1 | iexmacOS / Linux:bashcurl -fsSL https://raw.githubusercontent.com/KKKKhazix/agent-autoloop/main/install.sh | shAfter installation, run a trial:bashagent-autoloop try

License and pricing

The project is licensed under MIT. It is open source and free to use. Pricing is not stated beyond the open source license.

Based on the project's public README and repository metadata.

Add this badge to your README

Featured on YunoDB Badge styles & help

Maintain this project? Show that it is featured on YunoDB.

Markdown[![Featured on YunoDB](https://yunodb.com/badge/kkkkhazixagent-autoloop.svg)](https://yunodb.com/tools/kkkkhazixagent-autoloop?utm_source=github&utm_medium=badge)
HTML<a href="https://yunodb.com/tools/kkkkhazixagent-autoloop?utm_source=github&utm_medium=badge"><img src="https://yunodb.com/badge/kkkkhazixagent-autoloop.svg" alt="Featured on YunoDB" height="20"></a>