colemurray/background-agents — explained in plain English
Analysis updated 2026-05-18
Assign a background coding task to an AI agent from a Slack message or GitHub PR comment.
Set up scheduled or webhook-triggered coding tasks, like fixing errors flagged by Sentry.
Have multiple people join the same coding session to review an agent's work together in real time.
Split a large task into smaller pieces that run in parallel sandboxes.
| colemurray/background-agents | tutti-os/tutti | kentcdodds/kentcdodds.com | |
|---|---|---|---|
| Stars | 2,535 | 2,550 | 2,489 |
| Language | TypeScript | TypeScript | TypeScript |
| Last pushed | — | — | 2026-07-21 |
| Maintenance | — | — | Active |
| Setup difficulty | hard | easy | moderate |
| Complexity | 4/5 | 2/5 | 3/5 |
| Audience | developer | vibe coder | developer |
Figures from each repo's GitHub metadata at analysis time.
Requires setting up a GitHub App, a Cloudflare account, and a sandbox backend such as Modal, Daytona, or E2B.
Background Agents, also called Open-Inspect, is an open source system that lets a coding AI work on tasks in the background while you do something else. Instead of watching an AI type in your editor, you send it a task and it runs in its own sandboxed environment with access to Node.js, Python, git, and browser automation, then reports back or opens a pull request when it is done. You can start a session from several places: a web interface, Slack, a GitHub pull request comment, a Linear issue, or a webhook. Multiple people can join the same session at once, so a team can watch or steer an agent together in real time. When the agent finishes, it creates a pull request that is credited to the person who asked for the work, not a generic bot account. You can also set the agent to run on a schedule, such as a cron job, or to trigger automatically from a Sentry alert or an incoming webhook. For bigger jobs, the agent can break work into smaller pieces and run several of them at once in separate sandboxes. You can choose which AI model does the work, including Anthropic Claude, OpenAI Codex through a ChatGPT subscription, or OpenCode Zen. The project is built to run for a single organization rather than many separate customers. It uses one shared GitHub App installation to handle git operations like cloning and pushing code, and the README is explicit that it does not check whether an individual user is allowed to access a specific repository before starting a session. Pull requests made by users who sign in with GitHub use that user's own token, so they can only open PRs on repos they already have access to, but anyone who signs in another way falls back to the shared bot account. Because of this, the README recommends deploying it behind your organization's SSO or VPN, installing the GitHub App only on the specific repositories you want it to touch, and restricting who can sign in. Under the hood it runs on Cloudflare Workers and Durable Objects for coordination, with separate packages handling the web client, sandbox runtime, Slack and GitHub integrations, and infrastructure adapters for providers like Modal, Daytona, and E2B.
An open-source system that runs an AI coding agent in the background, working on tasks from Slack, GitHub, or Linear and opening pull requests when done.
Mainly TypeScript. The stack also includes TypeScript, Cloudflare Workers, Durable Objects.
No license file is mentioned in the README, so the terms of use are unclear.
Setup difficulty is rated hard, with roughly 1h+ to a first successful run.
Mainly developer.
This repo across BitVibe Labs
Verify against the repo before relying on details.