ColeMurray/background-agents

What it solves

Open-Inspect is a background coding system that allows developers to delegate software engineering tasks to AI agents that work independently in the background. It eliminates the need for developers to manually oversee every step of a coding task, enabling them to focus on other work while the agent handles implementation, testing, and PR creation across one or multiple repositories.

How it works

The system consists of a control plane (built on Cloudflare Workers) and a data plane consisting of isolated sandbox environments. When a task is triggered—via the Web UI, Slack, GitHub, or Linear—the control plane spawns a session sandbox containing a full development environment (Node.js, Python, git, etc.). The agent uses a runtime (OpenCode) to execute tasks, utilizing filesystem snapshots and pre-built images for near-instant startup. It can spawn parallel child sessions to decompose complex work and integrates with GitHub to create PRs with proper user attribution.

Who it’s for

It is designed for single-tenant organizational use where all users are trusted members of the same company and share access to the same repositories.

Highlights

• Multi-platform integration: Trigger agents via Slack, GitHub PR comments, Linear issues, or webhooks.
• Background Automation: Support for cron jobs and event-driven triggers (e.g., Sentry alerts or GitHub workflow completions).
• Multi-repo capabilities: Ability to work across up to 10 repositories simultaneously in a single session.
• Parallel Execution: Agents can spawn child sessions to handle sub-tasks in separate sandboxes.
• Flexible Model Support: Compatible with various providers including Anthropic, OpenAI, xAI, and OpenCode Zen.
• Full Dev Environment: Includes browser automation for UI verification and an optional browser-based VS Code instance.