Use case

Coding agents, run at fleet scale.

The moment you run more than one agent session at a time — one per ticket, one per branch, one per repo — you need per-session isolation, not a shared box the agents all fight over.

01

Autonomous bugfix loops

An agent reproduces a failing test, patches, reruns, and iterates until the suite is green or a budget is exhausted — the DeepSeek Harness pattern.

02

PR review & codegen bots

Triggered by a webhook, an agent checks out a branch, makes changes, and opens a pull request without a human ever provisioning a VM.

03

Parallel fan-out sessions

One sandbox per ticket, per branch, or per customer — dozens running concurrently, each with its own filesystem and process table.

04

Untrusted or external repos

Point an agent at a community PR or a repo you don't fully trust without extending it a lane onto your infrastructure.

In practice

Fan out one sandbox per unit of work.

No pool to pre-warm, no queue to manage — each create() call gets its own machine.

import ovrin client = ovrin.Client() # one sandbox per open ticket, fanned out concurrently for ticket in backlog: sandbox = client.sandboxes.create( template="claude-code", env={"ANTHROPIC_API_KEY": key}, ) sandbox.run(f"git clone {{ticket.repo}} /workspace/app") result = sandbox.run(f'claude "{{ticket.prompt}}"') open_pull_request(ticket, sandbox.run("git diff").stdout) sandbox.kill()

Pick your agent

Every major coding agent, same API.

Fan out your first ten sandboxes.

Same API whether it's one session or one thousand.

Create an API key Read the quickstart

$pip install ovrin