Hardware isolation
Every AI Computer is a dedicated micro virtual machine with its own kernel. Not a container. Not a namespace. A real machine.
Features
Serious infrastructure,
zero ceremony.
Every capability in Ostra answers one question: how do you let an AI run code instantly — without trusting it with your machine or your money?
Infrastructure
Built for the AI era.
The primitives you'd expect from serious infrastructure — reimagined for a world where software writes software.
Sub-second launch
From zero to a running computer in ~120 milliseconds. Faster than your terminal opens.
Financial guardrails
A hard budget on every computer, prepaid by design. At the cap, the machine dies and billing stops with it.
MCP & SDK native
pip install ostra for Python, npx @ostra/mcp for Claude and Cursor. Tools ship for LangChain and CrewAI.
Access anywhere
Terminal, editor, and preview in the browser — or expose any port and SSH in over IPv6 for raw access.
Disposable
Delete everything with one click. The whole machine vanishes — files, processes, secrets, mistakes.
Financial guardrails
Your agent gets a wallet.
Not your credit card.
Give your AI agent hands, legs, and a budget — without risking your cloud bill. Every computer runs inside a hard spending limit enforced by the platform, not by hope.
Hard caps, not alerts
Set a budget per computer. When it's hit, the microVM is destroyed and billing stops with it.
Prepaid by design
Agents spend from a wallet you fund. An autonomous loop can drain a budget, never your credit card.
Billed hourly
You pay for the hours a computer is alive. Destroy it, and the meter dies with it.
Agent wallet
computer running
$0.0000
spent this run · billed hourly
cap $5.00at cap → computer destroyed, spend stops
Free tier6 hrs · 1 computer at a time
Card on filenone — and none needed
agent.py — api.ostra.run
import ostra # one identity — OSTRA_API_KEY env works toowith ostra.Client(api_key="ostra_xxx") as client: result = client.deployments.create( project_id="proj-uuid", service_slug="postgres", plan_slug="small", environment="development", ) dep = result.deployment exp = client.deployments.expose(dep.id, port=5432) print(exp.endpoint) # public URL, live ssh = client.deployments.ssh(dep.id) print(ssh.ipv6.command)$ pip install ostra
$ npx @ostra/mcp ← for Claude & Cursor
MCP & SDK native
One SDK.
Full autonomy.
Deploy a service, expose a port, SSH in — three calls from Python. One command registers the MCP tool in Claude or Cursor. Your agent takes it from there.
Secure by default
MicroVM isolation on every deployment — not opt-in.
Budget on every computer
Hard spending caps, prepaid wallets, hourly billing.
Expose any port
One call turns a private port into a live public endpoint.
SSH built in
Every computer ships an IPv6 SSH command. Full access, no setup.
Projects & environments
development, staging, production — scoped out of the box.
Agent-native
MCP tool for Claude and Cursor. Python SDK for LangChain and CrewAI.
Use cases
One primitive. Endless workflows.
When environments cost nothing to create and nothing to destroy, whole categories of friction disappear.
agents
AI coding agents
Root access, a live URL, and a hard budget. Full autonomy, zero blast radius — financial or otherwise.
review
Code review
Actually run the PR before you approve it. Boot the branch on a fresh computer, click around, delete.
debug
Bug reproduction
A pristine machine for every repro. If it happens on a fresh computer, it's real — not your setup.
support
Customer support
Recreate a customer's exact environment in seconds instead of asking them for their Node version.
teach
Education
Every student gets an identical machine. No installation day, no 'it doesn't work on Windows'.
teams
Dev environments
Onboard engineers in minutes with template-based environments that are always up to date.
Give every AI a safe computer.
Don't let generated code touch your machine — or your card. Run it on a computer with a budget.
6 hours free · No card required · First computer in ~120ms