Skip to content
runcrate
Console

Runcrate Storage

Local-disk speed. Any GPU, any region.

Your files are on the instance's own disk when it boots, so jobs read and write at local-disk speed. Start the next GPU, in any region, and they're waiting.

Create a volumeRead the storage docs

10 GB to 10,000 GB a volume, grown any time.

One volume, carried to the next instance you launch, on any GPU.

What you get

Storage that follows the job. Not the machine.

  • Local-disk speed

    Your files are on the instance's own disk when it boots, and your job reads and writes them at local-disk speed.

    /workspace

  • Any GPU, any region

    Attach the volume to whichever GPU you launch next, wherever it runs. The instance's region never has to match the volume's.

    storage_id at launch

  • Files outlive the instance

    Terminate the instance and your work stays on the volume, with your latest changes, ready for the next GPU you start.

    kept after terminate

  • Browse files without SSH

    Open a volume in the dashboard to look through its folders and search them by name, with no instance running.

    dashboard · storage

  • Grow any time

    Add space as your data grows, up to 10,000 GB a volume. The days you prepaid at the old size come back.

    volumes resize

  • Your bucket, 39 regions

    The files sit in a bucket in a storage account you connect, in the US, Europe, Asia Pacific, the Middle East, South America and more.

    an account you own

The same balance and API key pay for volumes, instances and model calls, and your team shares the workspace's volumes.

How it works

Start a GPU. Your files are already there.

Nothing to download when a job starts and nothing to save when it ends. Every instance you launch with the volume has your files.

Using a volume on an instance
  1. 01

    Create a volume.

    Pick a name, a size and a region, in the dashboard or with runcrate volumes create.

  2. 02

    Launch any GPU with it.

    Choose the volume in the launch form, or pass storage_id from the SDKs or the API. Any GPU, in any region.

  3. 03

    Your files are just there.

    In /workspace, on the instance's own disk, read and written at local-disk speed.

  4. 04

    Stop it. Start another.

    Terminate the instance whenever you like. The next GPU you start with the volume has the same files, where you left them.

Examples

Ask your agent. Or read the code it writes.

Claude Code or Codex makes the volume with the runcrate CLI, then launches the GPU with it from the SDK.

Docs

Persistent cloud storage volumes that survive instance termination.

Make a 100 GB volume called datasets in us-east-1, then launch an H100 with it attached using the Runcrate SDK. Runs runcrate volumes create --name datasets --size 100 --region us-east-1. It returns: the volume, created. Runs runcrate volumes list --json. It returns: the volume, with its id. Runs npm install @runcrate/sdk. It returns: the TypeScript SDK, installed. Writes launch.mts: the launch from the SDK docs, with storageId set to the volume. Runs npx tsx launch.mts. It returns: the H100 queued, with the volume attached.

Pricing

$0.03 per GB a month. For the size you pick.

Volume

What it costs

  • 10 GB

    The smallest volume

    $0.30 a month

  • 100 GB

    $3.00 a month

  • 1,000 GB

    $30.00 a month

  • 10,000 GB

    The largest volume

    $300.00 a month

  • Prepaid a week at a time from your dollar balance, starting the day you create the volume.

  • Delete a volume or resize it and the unused days come back to your balance.

  • A volume bills while it exists, attached or not. The instances you attach it to bill separately.

  • If your balance can't cover the next week, the volume is suspended, and can't be attached, until it can.

See every price

Options

Four fields. Three to create, one to attach.

Name itname
Default: required
How the volume shows in the dashboard, the launch form and the CLI.
Pick the sizesize_gb
Default: required
From 10 to 10,000 GB. Grow it later; it never shrinks.
Pick the regionregion
Default: required
Where the bucket lives, from your storage account's regions. runcrate volumes regions lists them.
Attach it at launchstorage_id
Default: none
On the instance you create: storageId in TypeScript, storage_id in Python and the API, or the Volume field in the launch form. The CLI creates volumes but doesn't attach them; over MCP, create_instance takes storage_id.

Where a volume stops

  • One running instance at a time, attached as it launches. A second launch with the same volume is refused until the first is terminated.

  • A volume belongs to one environment in your workspace, and attaches to instances launched there.

  • A volume with more files takes longer to be ready when the instance boots.

  • It is your folder, not a backup: a file you delete on the instance is gone from the volume too. node_modules folders aren't kept.

Read the storage docs

FAQ

Questions about volumes.