borrowed
machines Request early access
Compute for the things you're building.BM / 001

Big ideas.
Borrowed machines.

A place for your AI
to get to work.

Run code in isolated sandboxes. Keep your files. Branch your next idea.

See how it works
Your intelligence. Our infrastructure.Isolated by design.
01 / The machine

Great minds need
a place to run.

Your model can write the code.
Now give it somewhere to execute.

Borrowed Machines is the execution layer for AI applications. Send code through an API, run it in a Firecracker microVM, and get the results back. Bring your own model, framework, and ideas.

[ 01 ]

Room to experiment.

Run generated code inside an isolated virtual machine, with execution limits you control.

MicroVM isolation
[ 02 ]

Work worth keeping.

Keep files in persistent workspaces. Start from a snapshot, fork a new branch, and build on what worked.

Persistent workspaces
[ 03 ]

Connections with limits.

Scope outbound API access and bind only the credentials an execution needs.

Controlled network access
02 / In practiceInteractive walkthrough

A little code.
A lot of possibility.

Execute. Keep the result.
Take the next idea for a spin.

Python / main.py
from statistics import mean

measurements = [18, 24, 21, 29]
result = mean(measurements)

print(f"Average: {result}")
Runs inside an isolated microVM
Sample outputReady
Average: 23

Send code, then retrieve its output and execution status through the API.

execution / completed

An illustrative walkthrough. No code is sent to a live sandbox from this page.

03 / Bring your own ambition

You bring
the “what if.”

01

Agents that do things.

Give an agent a place to execute code, inspect files, and turn a plan into a result.

02

Data with an answer.

Process a dataset, run an analysis, and keep the output for the next step.

03

Ideas with a second draft.

Branch a workspace to test a different approach without changing the original.

One execution layer.Your choice of model. Your choice of framework.
04 / A few details

Good questions.

Is this an AI model or an agent framework?

It is the infrastructure that runs code. Your application decides what to do; Borrowed Machines provides the sandbox. Use it with your own model, agent framework, or an ordinary API client.

What happens to files after an execution?

Choose a transient execution for scratch work, or attach a persistent workspace to keep files. Workspace snapshots let you start later work from a known state.

Can code connect to external services?

You can grant scoped outbound access and bind credentials for the services an execution needs. Access is part of the execution configuration.

How do I get started?

Request early access and tell us what you want to build. We will discuss access and the right setup for your workload.

Have something in mind?

Let's put it
to work.

Building with AI?
Request early access to Borrowed Machines.

Request early access