Introducing Render Sandboxes: a computer for every agent.

Render Sandboxes start fast, can run indefinitely, and scale reliably to thousands at once. Each one is a fully isolated microVM with configurable access to the internet.

Idle sandboxes suspend automatically by default, then restore with files and memory intact. You don’t pay for CPU or memory while one is suspended, and active sandboxes are never billed for unused CPU.

Run any agent or harness in a sandbox, right alongside the rest of your stack on Render. Read the launch post:

🎉 congrats team! also a great video!!

Agents deserve their own computer, lets go!! 🚀🚀

🚀

SO EXCITING

Thrilled to see this go live to all our developers! So much work went into this one across engineering, product, and marketing, and I couldn't be more proud

at this rate my local coffee shop offers sandboxes too by next week

@lanreadelowo

Computer for every agent. Let's go

Very cool launch!

For approval-heavy agents, RAM-hours may track human response time more than model speed. That makes the point where you suspend a workflow decision.

Render Sandboxes: isolated runs and resuming saved state 1/2 I explained this as a manga.

A computer for every agent

one sandbox per agent needs a spending cap when failed commands trigger repeated retries

@lanreadelowo

@lanreadelowo

给每个智能体单独配台电脑，至少不用把所有任务都挤在同一个工作环境里了。

A computer for every agent is such a clean framing — sandboxes that feel like a real machine, not another toy REPL.

A dedicated computer for every agent is brilliant

Why is everyone making sandboxes for agents

suspend with memory intact is the real feature. a lot of agent wall time is just waiting on the model, not burning cpu