Managed sandboxes, one API

We run the Firecracker capacity, the database, object storage, TLS, backups and upgrades. You create sandboxes, run code and destroy them.

Get an API key Read the docs

What you get

Firecracker isolation by default

Public workloads run in microVMs with their own kernel and filesystem. You cannot request a weaker runtime and downgrade your own isolation.

Disposable by default

Create, use, destroy. Sandboxes are short-lived and billed while they run, so agents can churn through them freely.

Coding images

Git, CA certificates, Python, curl and tar are in the image. Clone over HTTPS with real certificate validation, edit in place, run the tests.

Works with your agent

The API is the product. Use the Python SDK, the Rust SDK, or plain HTTP from whatever runtime your agent is written in.

How it fits together

your agent
  -> AIec API        auth, tenants, quotas, policy
  -> scheduler             fenced leases, capability-aware placement
  -> worker                HTTPS control channel
  -> runtime               Firecracker microVM, or hosted capacity
  -> sandbox               your code, isolated

Using it

from agentforge import AIec

af = AIec(api_key="af_live_...")

box = af.sandboxes.create(image="python:3.13")
box.exec(["python", "-c", "print('hello from AIec')"])
box.destroy()

Prefer to run the platform yourself instead? It is the same software: self-host AIec.

Alpha status

Cloud is in public alpha: invite-only, limited capacity, and no SLA. Current limitations are listed on the status page. Nothing here bills without your approval.