runtm-ai/runtm

What it solves

Runtm provides secure, isolated environments (sandboxes) where AI coding agents can write, test, and deploy software without risking the user's local machine or cloud infrastructure. It eliminates the risk of "yolo agents" executing dangerous code by providing OS-level isolation and a streamlined path from development to live deployment.

How it works

Users can spin up a sandbox session using a CLI tool. These environments use bubblewrap (on Linux) or seatbelt (on macOS) for fast, secure OS-level isolation without the need for full containers. AI agents (such as Claude Code, Cursor, or Codex) operate within these sandboxes to build software. Once the code is ready, a single command deploys the project to a live HTTPS endpoint on auto-stopping infrastructure that wakes up on traffic.

Who it’s for

Developers and AI agent users who want to safely execute AI-generated code and instantly deploy the results to the web.

Highlights

• OS-level isolation: Fast startup (<100ms) using native OS sandboxing tools.
• Multi-agent support: Compatible with various coding agents including Claude Code, Codex, and Gemini CLI.
• Instant deployment: One-command deployment to live HTTPS URLs with logs and previews.
• Persistent workspaces: Ability to stop and resume sessions while preserving files.
• Self-hosting: Option to fully self-host the entire infrastructure.