Freestyle
Freestyle is the infrastructure for code you didn't write — VMs and Git for AI agents. The platform provides Linux microVMs that boot in under 600ms with live-fork, pause-resume, and persistent snapshots; a multi-tenant Git service with branchable filesystems, GitHub Sync, full-text search, and webhook triggers; an Execute (Serverless Runs) API for ephemeral JavaScript/TypeScript code; a Web Deployments API for hosted Node.js apps and static sites; Cron schedules; custom Domains with wildcard SSL; an Identity service for scoped per-user and per-agent access tokens; and Observability logs. Freestyle is a direct sandbox option for AI app builders, background agents, code review bots, and long-running assistants — a usage-priced alternative to running Anthropic's hosted Code Execution tool or building bespoke microVM infrastructure on Firecracker, Modal, or E2B.
8 APIs
21 Features
AIAgentsSandboxesVMsMicroVMsGitCode ExecutionJavaScriptTypeScriptServerlessHostingDeveloper ToolsInfrastructure
Create, snapshot, fork, suspend, resume, exec, and manage Linux microVMs designed for AI agents. VMs boot in under 600ms with restored memory snapshots, support live-forking a r...
Multi-tenant hosted Git for AI agents — every agent gets a branchable filesystem with commits, diffs, rollback, and review. Includes create/list/delete repositories, commits fro...
Identity and access management for end users and AI agents. Create identities, mint and revoke access tokens, grant per-repo Git permissions (read/write/admin), grant per-VM per...
Manage custom domains for Freestyle Web Deployments and VMs — list domains, create/verify ownership, insert and remove domain → deployment mappings, provision wildcard SSL certi...
Serverless Runs — execute JavaScript or TypeScript code on demand and get the result. No deployment, no HTTP server — POST code and receive the output. Supports node modules, en...
Deploy and manage Node.js web applications and static sites. Push code via /web/v1/deployment, list deployments, inspect deployment metadata, and fetch/update files on a deploym...
Schedule recurring serverless runs. Create, update, and delete cron schedules; list past executions; inspect per-schedule success rate and a metrics timeline for observability.
Query structured logs across Freestyle services. Filter by VM, deployment, run, identity, or time range to debug agent runs end-to-end.
Linux microVMs that boot in under 600ms with restored memory snapshots
Live VM forking — clone a running VM into multiple copies in milliseconds
Pause and resume — hibernate idle VMs and pay nothing while paused
Snapshots (memory + disk) with persistent snapshot retention on Hobby+
Multi-tenant Git hosting with branchable filesystems for AI agents
GitHub bidirectional sync (pull, push, or both) per repository
Full-text search across files, commit messages, and diffs
Webhook triggers on commits and refs
Per-repo and per-VM identity permissions with scoped access tokens
Custom domains with verification, wildcard SSL, DNS record management, and deployment mappings
Serverless Runs (execute) for ephemeral JS/TS code execution
Serverless Deployments for hosted Node.js apps and static sites
Cron schedules with execution history and metrics timeline
Observability logs across VMs, deployments, runs, and identities
First-class language integrations — Node.js, Bun, Deno, Python, uv, Ruby, Java, PostgreSQL, OpenCode, Web Terminal
TypeScript and Python SDKs plus Vercel AI, LangGraph, and Mastra SDK adapters
rigkit CLI for local dev machines
SSH access to VMs
Bearer-token auth (Authorization header) for all REST APIs
Pricing per vCPU-hour, GiB-memory-hour, and GiB-storage-hour with daily free allowances
Plan tiers Free, Hobby ($50/mo + usage), Pro ($500/mo + usage), Enterprise
AI app builders (Lovable, Bolt, V0 style) provisioning a sandbox VM per project with cloned source, dev servers, and *.style.dev preview domains
Background agents (Devin, Cursor Agent style) forking a base VM into parallel workers — one builds the API, one builds the UI, one writes tests
Code review bots (CodeRabbit, Greptile style) cloning a repo at a PR SHA, running lint/test, and posting a Claude-generated review
Long-running AI assistants (Claude, OpenClaw, Cowork style) using a persistent VM with 60s idle pause to keep per-user state at $0 cost between turns
LLM code-interpreter style — execute model-generated code safely in an isolated microVM with egress control
Sandboxed AI agent evals on ephemeral microVMs (one VM per eval run)
Multi-tenant SaaS where each end user gets their own Freestyle Git repo and isolated VM
Reinforcement learning on microVMs — fan out, train, fold in
Deep-research agents that spawn, suspend on idle, and resume on event
Hosted code playgrounds for educational and developer-tool products
Anthropic Claude — Code Execution tool, Managed Agents, Skills (Freestyle sandbox positioned as an alternative or complement to Anthropic's hosted code execution sandbox)
GitHub — bidirectional repo sync, GitHub Sync configuration per repo
Vercel AI SDK — first-party adapter in sandbox_sdks
LangGraph — first-party adapter in sandbox_sdks
Mastra — first-party adapter in sandbox_sdks
Node.js, Bun, Deno — first-class VM runtimes
Python (with uv), Ruby, Java — first-class VM runtimes
PostgreSQL — managed inside VMs
OpenCode — VM integration
Cloudstate — Freestyle's own JavaScript database runtime
Adorable — Freestyle's own open-source Lovable-style AI app builder
Astro, Next.js, SvelteKit, Solid Start, React Native, Expo, Vite — starter templates
