Managed sandbox cloud

Sandboxes for AI agents & RL, at fleet scale

Collimate runs your AI-generated and agentic code in fast, isolated microVMs — then lets you fork running state, scale to thousands of environments, and suspend and resume on demand. Through a clean API, SDK, and MCP server.

Get started View pricing

# Spin up an isolated sandbox and run code in it from collimate import Collimate sbx = Collimate().sandboxes.create(template="python") print(sbx.exec("print(6 * 7)").stdout) # → 42

What you can do

Fork · Scale · Store

The primitives agentic and reinforcement-learning workloads actually need — built into the platform, not bolted on.

⑂

Fork running state

Clone a live sandbox — memory, processes, and disk — into many isolated children in milliseconds. Explore branches, run rollouts, and backtrack without re-running setup.

≡

Scale to a fleet

Schedule thousands of concurrent sandboxes across a managed fleet, with per-tenant isolation, quotas, and usage metering built in. Pay for exactly what you run.

⤓

Store & resume

Suspend an idle sandbox to disk and resume it later exactly where it left off. Metering pauses while it sleeps, so long-running agents stay cheap.

microVM

Hardware-isolated, per-sandbox — not shared containers

API · SDK · MCP

Drop into any agent framework or RL trainer

Usage-based

Per-second compute, storage, and snapshots — suspend stops the meter

Get started

Run your first sandbox today

Start free on the Demo tier — no card required. Move to Pro when you need scale.

Create an account See pricing