Today's AI open source · 2026-11-12
KKKKhazix/agent-autoloop
Agent Autoloop is an automation tool that lets command-line AI agents take on tasks in the background and carry them through to completion.
| Repository | github.com/KKKKhazix/agent-autoloop ↗ |
|---|---|
| Main language | TypeScript |
| License | MIT |
| GitHub stars · forks | ★ 4,526 · 827 forks (as of 2026-11-12) |
| Created · last update | 2026-11-03 · 2026-11-11 |
| GitHub topics | ai-agents, automation, claude-code, cli, codex, event-driven, mcp, nodejs |
Overview
Agent Autoloop is a lightweight automation relay that lets command-line agents such as Claude Code and Codex run without a person sitting at the terminal watching. When a scheduled time arrives or an external event comes in, it hands the task to an agent, and only for tasks that need judgment does it send a card to a person to get a decision.
The internal core that handles task scheduling has no external dependencies, and event sources, agents, notification channels and deployment pipelines are all swappable plugins. Users write in rules.json 'which task, to which agent, in which directory,' and the agent appends an agent-autoloop code block to the end of its final response to leave questions, notifications or follow-up check reservations. While waiting for a decision, it keeps no process running and uses no tokens; once it gets an answer, it resumes the original conversation and finishes the remaining work.
It suits developers who already use command-line agents and want to hand off recurring work such as responding to issues, writing morning reports or checking outage alerts. It runs on a single personal computer, without a cloud server or public ports.
Why it is worth a look
A notable design choice is that it lets command-line agents act as background workers that receive events and do work, rather than as interactive tools. It implements a Human-in-the-loop flow — asking a person only for risky tasks, using no resources while waiting, then returning to the same conversation to continue — through a single response format, without an SDK or changes to agent code.
Key features
- Assigns events from schedules, GitHub/GitLab Webhooks and HTTP requests to designated agents and directories according to rules.
- Supports Claude Code, Codex, ACP, DeepSeek Harness and ZCode as built-in plugins, and any command-line tool that takes a prompt and outputs a final response can be connected with one line of configuration.
- Tasks that need judgment, such as deleting data, restarting services, spending money or sending external messages, trigger a card via desktop notification, Feishu, DingTalk, WeCom, Slack or Discord to get a decision.
- Provides commands to pause and resume rules, replay missed events, view run history and details, and check and dry-run rules.
- A deployment plugin carries a PR through review, waiting for CI, merging and deployment verification once a deliver label is added.
What it is used for
- Have it compile local notes and Git commits into a report every morning.
- Periodically check new GitHub issues, fix the code locally and get notified of the results.
- Receive outage alerts, compare them against the code to find the cause, and proceed with a restart only after confirmation if one is needed.
- Check website status, or sort unread emails and draft replies.
Getting started
- Prepare Node 24 or later, Git and a logged-in command-line agent (macOS and Linux also need curl).
- Install with curl -fsSL https://raw.githubusercontent.com/KKKKhazix/agent-autoloop/main/install.sh | sh on macOS and Linux, or irm https://raw.githubusercontent.com/KKKKhazix/agent-autoloop/main/install.ps1 | iex in Windows PowerShell.
- Run agent-autoloop try to go through the question, decision and continuation flow once.
- Add rules to ~/.agent-autoloop/rules.json, then check them with agent-autoloop check.
What to weigh
- The project was created in November 2026, so the rule format and commands may change.
- According to the README, the scenario templates were verified only with local HTTP and a mock Feishu CLI, and full workflows have not been run on real accounts.
- Because agents work on their own in the background, you should set the token budget and permission scope of the agent you use in advance.
- Some templates default to Beijing time and Feishu direct messages, so they need to be adjusted for your environment.
Related terms and companies
Sources
Written with AI assistance from the repository README and GitHub data. Star counts and other numbers are GitHub values as of the date shown; features and licensing can change, so check the repository.