Launch HN: Freestyle – Sandboxes for Coding Agents

Freestyle has launched a new infrastructure platform designed specifically for running AI coding agents at scale. The service provides isolated sandbox environments where developers can deploy agents like those powering tools such as Cursor, Devin, or Claude Code. Key technical specs include virtual machines that provision in under 700ms, live forking capabilities that clone running VMs in milliseconds, and persistent storage that allows agents to maintain state across sessions while costing nothing when paused.

The platform supports three primary use cases demonstrated in their launch examples. First, developers can create app-building workflows similar to Lovable or V0, where agents run development servers and handle real-time coding tasks. Second, teams can implement parallel development patterns where multiple agent forks work simultaneously on different code components—like having one agent build API endpoints while another builds frontend UI. Third, the infrastructure enables automated code review systems similar to Code Rabbit, where agents can run linters, execute tests, and provide pull request feedback.

Freestyle's architecture addresses the scaling challenges of AI agent deployment. Traditional cloud infrastructure wasn't built for the rapid provisioning, ephemeral workloads, and state management requirements of AI agents. By offering specialized sandboxes with instant startup, zero-cost pausing, and millisecond cloning, Freestyle enables developers to run what they describe as 'tens of thousands of agents' simultaneously. The platform integrates directly with Git repositories and provides programmatic control through a TypeScript SDK, making it accessible to developers already working with modern AI coding tools.