Volumes

A Volume is /workspace. Your code, your checkpoints, your datasets. What you made, not what you installed. If losing it would hurt, it belongs here.

A volume is standalone and team-owned, not tied to one pod. Any number of your pods can point at the same volume; only one of them can be running with it at a time. It restores to the box's own local disk at Start, so a running pod reads and writes at local-disk speed with the filesystem semantics that pip, git, and databases actually depend on, then saves back to object storage while you work and at Stop.

Volume and environment are independent

A pod's Environment is the software it starts from: the base image, what you installed, /opt, /root. A pod's Volume is its data. Either exists without the other, and deleting one never breaks the other. What a Stop keeps depends on which ones your pod has:

ComfyUI pods keep models, inputs, outputs, and workflows on the pod's volume the same way, so they come back with the rest of /workspace on the next Start.

Only one running pod at a time

Any number of pods can be pointed at the same volume, and a stopped pod never blocks anything: another pod can pick the same volume, roll it back, or the volume can be deleted out from under it (a stopped pod using a deleted volume just starts with an empty /workspace next time). The one rule is that only one running pod writes a volume at a time.

That rule is enforced at Start, before anything is rented. Starting a second pod on a volume already in use is refused up front:

Volume "training-data" is in use by pod "finetune-run-3". Stop it first.

Status

A volume's status is one of:

Automatic history

There is no manual save button. Every Stop of a pod using a volume writes one new point in its history, labelled by why it happened:

The volume's head is what the next Start restores. Stop and a successful save tick both move it forward; nothing ever moves it backward except you, deliberately, with Restore a point.

Saved while the pod runs

Between stops, a volume saves itself in the background so a sudden provider kill loses only the interval since the last save, not everything since the last Stop.

These background saves do not create new history points. They exist purely to shrink how much a provider kill can cost you; Restore a point only ever offers you the points written by an actual Stop.

Restore a point

Pick any point in a volume's history and set it as the head. This is only available while the volume is not in use by a running pod: stop that pod first.

aq volume restore <id> <point-id>

Restoring never deletes later points from the history; it only changes what the next Start reads.

Attach to a pod

From a volume's row, Attach to pod opens a picker of your team's pods: any stopped pod, or New pod. Picking a stopped pod takes effect the next time it starts.

A running pod's own Storage section has Change and Detach for its volume. Both need the pod stopped first, and take effect at its next Start:

• Change swaps in a different volume of yours.
• Detach removes the volume; the pod starts with a clean, empty /workspace until you attach another one.

There is no aq subcommand yet for changing or detaching a pod's volume; both are console-only.

Duplicate

Duplicating a volume makes an independent copy right away: a new volume, its own history, seeded from the source's current head. Writes on the copy never flow back to the source, and later writes on the source never appear on the copy.

aq volume dup <id> <new-name>

This is the move for "start a second pod from the same data without the two ever writing over each other."

Delete

aq volume rm <id>            # delete; refused while in use

Delete is refused only while a running pod is using the volume; stop it first. Deleting a volume that stopped pods still point at is allowed: those pods are detached and start with an empty /workspace next time. Deleting a volume deletes its whole history; there is no per-point delete.

Deleting a pod does not delete its volume by default. Tick Also delete volume if you want both gone together; otherwise the volume is kept, still billing, for you to reattach or clean up later.

Bringing in a volume from outside

aq import [flags]

Adopts data you already have (from a box you rent elsewhere) as a new volume, without going through a pod first. See aq import.

Sharing

Sharing an environment (see Sharing an environment) can optionally include files: a frozen copy of one saved point on your volume, at the time you share. The recipient never gets your live volume; they get a new volume of their own, duplicated from that point.

What it costs

You pay to hold a volume's history at $0.02/GiB per month. Saving while you work is free; only the bytes held count. Deleting a volume you no longer need is the way to bring that number down; there is no per-point delete, so a volume's whole history goes together.