Freestyle Launches Sandboxes for AI Coding Agents with Live Forking

What Freestyle Offers

Freestyle is building cloud infrastructure specifically for AI coding agents, providing sandboxes that function as full virtual machines. These VMs are designed to be interchangeable with EC2 instances from an agent's perspective, but with specialized features for AI development workflows.

Key Technical Features

• Live Forking: Can fork a running sandbox horizontally with less than 400ms pause. This forks the entire memory state, not just the filesystem. If you're halfway through a browser page with animations, running a Minecraft server, or have an error in process, all forks will maintain that exact state.
• Fast Startup: Sandboxes start in approximately 500ms, with demos showing VM provisioning in under 700ms from API request to ready machine.
• Full System Support: Runs full Debian with hardware virtualization, supporting eBPF, Fuse, systemd init instead of runc, and multiple users. The goal is that anything expected to work on Debian should work on these VMs.
• Snapshotting: Can save VM state and resume weeks later from the exact point.
• Persistence Options: Supports persistent VMs that pause after idle timeout (e.g., 60 seconds) with $0 cost while paused, resuming on next execution.

Infrastructure Approach

Freestyle runs on their own bare metal racks after finding that moving VMs across cloud nodes didn't provide acceptable performance. They discovered that monthly costs for Google Cloud and AWS bare metal nodes were equivalent to the total hardware cost, leading them to build their own infrastructure.

API Usage Examples

The source shows several code patterns for different use cases:

// App Builder pattern (like Lovable, Bolt, V0)
import { freestyle, VmSpec } from "freestyle-sandboxes";
import { VmBun } from "@freestyle-sh/with-bun";
import { VmDevServer } from "@freestyle-sh/with-dev-server";

const { repoId } = await freestyle.git.repos.create({ ... });
const { vm } = await freestyle.vms.create({
  with: {
    devServer: new VmDevServer({
      devCommand: "bun run dev",
      runtime: new VmBun(),
      repo: repoId
    }),
  },
});

// Agent pattern (like Devin, Cursor Agent)
import { freestyle, VmSpec } from "freestyle-sandboxes";
import { VmBun } from "@freestyle-sh/with-bun";

const { vm } = await freestyle.vms.create({
  git: {
    repos: [
      { repo: "https://github.com/user/repo.git" },
    ]
  }
});
const { forks } = await vm.fork({ count: 3 });
await Promise.all([
  ai(forks[0], "Build the API endpoints"),
  ai(forks[1], "Build the frontend UI"),
  ai(forks[2], "Write the test suite"),
]);

// Code review pattern (like Code Rabbit, Greptile)
import { freestyle } from "freestyle-sandboxes";
import { VmBun } from "@freestyle-sh/with-bun";

const { vm } = await freestyle.vms.create({
  git: {
    repos: [{ repo: repoUrl, rev: branchRev }],
  },
});
const { stdout: lint } = await vm.exec("bun run lint");
const { stdout: test } = await vm.exec("bun test");
const review = await ai(vm, "Review the diff for bugs");
await github.pulls.createReview({
  body: review,
  event: test.includes("FAIL") ? "REQUEST_CHANGES" : "APPROVE",
});

Target Audience

This infrastructure is designed for developers building or using AI coding agents that need full-system sandboxes for testing, development, and deployment workflows at scale.