Freestyle provides high-performance infrastructure primitives specifically engineered for AI agents, offering full Linux virtual machines and Git-backed filesystems. By moving away from restrictive sandbox-shaped containers, the platform grants developers access to real Linux environments with full root access, systemd support, and a complete networking stack, enabling agents to run complex applications, install packages, and manage persistent services effectively.

Freestyle’s VM architecture is designed for speed and scale, capable of provisioning virtual machines in under 600ms. It supports advanced features such as live forking, which allows users to clone active VMs in milliseconds, and a robust pause-and-resume mechanism that lets agents hibernate and resume their exact state, ensuring developers only pay for compute resources when they are actively used. This capability enables workflows ranging from parallel feature exploration to background processing and persistent state management.

In addition to compute, Freestyle provides an API-first Git filesystem platform that treats Git as a fundamental component of the agent's workspace. This system enables developers to manage agent-generated state through branches, commits, diffs, and rollbacks. It supports granular access control, webhooks, and bidirectional synchronization with GitHub, ensuring that code or agent output remains tracked, auditable, and easily shared across development teams.

Some of the key features are:

• Instant VM Startup: Provision full Linux VMs from API request to ready-to-run state in under 600ms.
• Live Forking: Clone running virtual machines instantly without pausing, facilitating parallel processing and branching workflows.
• Pause and Resume: Hibernate VMs to freeze CPU and memory, stopping compute billing while preserving all state for later resumption.
• Full Root Access: Deploy applications with real root privileges, including support for Docker, nested virtualization, and full networking stacks.
• Branchable Filesystems: Use Git as the primary filesystem for agents, providing native version control, diffing, and rollback capabilities.
• API-First Design: Manage all infrastructure resources, including repositories and VMs, programmatically via a comprehensive REST API and SDKs.
• GitHub Synchronization: Keep Freestyle Git repositories in sync with GitHub with bidirectional updates.

Freestyle operates by providing a programmable infrastructure layer that integrates directly into AI agent workflows. Developers interact with the platform through a TypeScript SDK or the Freestyle CLI, enabling them to automate the creation of VMs, manage repository lifecycle events, and handle complex stateful environments without the overhead of manual infrastructure management.

Some common use cases include:

• AI Coding Agents: Give each agent its own isolated environment with the ability to fork VMs to explore multiple solutions in parallel.
• Agent Filesystems: Build durable, versioned filesystems for customer tasks that require collaborative review and audit trails.
• Browser Automation: Run headless browsers within full Linux VMs for testing, web scraping, or AI-driven interaction tasks.
• Reinforcement Learning: Create golden environments that can be fanned out into thousands of parallel rollouts to train AI agents at scale.
• Application Previews: Utilize Git webhooks and VM snapshots to build automated preview environments for every push.