# What is a sandbox fork?

A sandbox fork is a new sandbox copied from a running sandbox. The copy has the
same files, memory and running processes as the original at the moment of the
copy.

**On Runtime, one call makes 1 to 100 running copies. The snapshot
that a fork takes for itself is free.** Copies are billed as new sandboxes of
the same size. A 2 vCPU, 4 GiB copy costs $0.03125 an hour while it waits,
and $0.08 an hour with both CPUs busy ([pricing](/docs/pricing)).

## Why AI agents fork

An agent often reaches a point worth branching from. The repository is cloned,
the dependencies are installed, the tests have run once, and a server is up.
From there, the agent may want to try three fixes, sample several model
answers, or run many evaluation episodes. A fork gives each attempt its own
copy of that moment, including its memory and processes. No attempt repeats the
setup, and no attempt can disturb another.

A fork differs from a fresh sandbox built from an image. An image holds files.
A fork also holds what was in memory: a loaded model, a warm cache, a shell
partway through a session, and a running dev server.

## Runtime's forks in facts

| Fact                | Value                                                                                                                |
| ------------------- | -------------------------------------------------------------------------------------------------------------------- |
| What is copied      | Files, memory and running processes                                                                                  |
| Copies per call     | 1 to 100 on prepaid credit, 1 to 4 without credit, all running when the call returns |
| Size                | Each copy gets the source's vCPUs, memory, disk and CPU setting                                                      |
| Price               | Each copy is billed as a create of that size would be                                                                |
| The fork's snapshot | Deleted when the fork ends and not billed, unless you keep it                                                        |
| Source sandbox      | Paused for the capture, then woken before the fork returns                                                           |
| Where copies run    | On the source's server while it has room; on another, the snapshot copied there first, when it is full               |
| Funding             | Accepted and ignored: copies run on what the account runs on                                                         |

If one copy fails, the error's
`details.startedSandboxIds` names the copies that did start
([snapshots and forks](/docs/javascript#snapshots-and-forks)).

## Fork a prepared sandbox

```ts check
import { Sandbox } from "withruntime";

await using base = await Sandbox.create();
await base.exec("pip install --quiet requests");
const [a, b] = await base.fork({ count: 2 }); // both running, answered together
await Promise.all([a!.stop(), b!.stop()]);
```

```python check
from withruntime import Runtime

runtime = Runtime()
with runtime.sandboxes.create() as base:
    base.exec("pip install --quiet requests")
    for fork in base.fork(count=2):
        fork.stop()
```

The CLI does the same with `runtime sandboxes fork <id> --count 3`. Pass
`keepSnapshot: true` to keep the fork's snapshot and start more copies later.
The kept snapshot is then billed as [snapshot](/glossary/sandbox-snapshot)
storage, at $0.08 per decimal GB per 30-day month (rate in
force since 23 September 2026). An agent connected through MCP forks with the
`runtime_sandboxes_fork` tool.

## Related

- [What is a sandbox snapshot?](/glossary/sandbox-snapshot)
- [RL environments](/use-cases/rl-environments)
- [Agent evals and SWE-bench](/use-cases/agent-evals-and-swe-bench)
- [A sandbox for a coding agent](/use-cases/coding-agent-sandbox)
- [Pricing](/docs/pricing#snapshots-images-and-volumes)

Facts on this page were checked on 25 October 2026.