Storage

Pods are disposable; volumes are not. A persistent volume keeps your data across pod stops, restarts, and re-creations. Meshive has two kinds:

	Local volume	Network volume
Lives on	The pod’s own host machine (fast, direct disk)	A dedicated storage node (NFS)
Created	Together with a pod, in the create flow	Standalone, from Resources › Storage
Tied to	That one machine — the pod must stay there	Any pod in the workspace can mount it
Pod types	Demand pods only	Demand and spot
Encryption	Opt-in at creation (details)	Opt-in at creation (details)
Pod outputs to Assets	Harvested from watched folders	Not harvested — you manage the files yourself

Sizes start at 10 GiB per volume, up to what the host has free. Rates are per GiB·month — see Pricing & Instances.

Note

Storage is billed while the data exists, including when the pod is stopped — the bytes still occupy a physical disk. The exception is host maintenance: a volume on a machine with an open maintenance window isn’t billed until the window ends. That includes a network volume on a storage node in maintenance — the volume isn’t billed, but pods on other machines that mount it still are. Delete volumes you no longer need. When you delete a pod, the confirmation dialog can also delete its attached local volumes in the same step; network volumes stay and are deleted from the Storage tab.

Connect — access without a pod

A volume doesn’t need a running pod to be reachable. Connect on the volume card opens a small access channel with SSH credentials (terminal and web) so you can upload or pull data directly. Preparing the channel usually takes about a minute.

For encrypted volumes, Connect is also what unlocks the volume — and an idle countdown locks it again automatically.

What the channel is for

Good for	Use a pod instead
Uploading and downloading files	Training and inference
Fetching models and datasets — wget, curl, aria2c, hf download, git clone	Builds and compiling
Unpacking archives	Installing large packages

Limit
CPU	0.5 vCPU
Memory	2 GiB
GPU	None
Space outside the volume	1 GiB — your home folder (/root), /tmp, and anything you install with apt or pip

The volume is mounted at /mnt/<volume-name>, and SSH sessions start there. The Connect modal shows the exact path — if you rename a volume, a channel that’s already running keeps the old path until it’s recreated.

Caution

If memory or the 1 GiB outside the volume runs out, the channel restarts: open sessions disconnect and everything outside the volume (installed tools, files in /root or /tmp) is removed. Files in the volume are always kept. The volume card and the Connect modal show when it happened and why.

Download straight into the volume

wget -P /mnt/<volume-name> <url>
aria2c -x 16 -d /mnt/<volume-name> <url>
hf download <repo> --local-dir /mnt/<volume-name>/<name>
git clone <repo-url> /mnt/<volume-name>/<name>

Caches and temporary files (Hugging Face, pip) go to /mnt/<volume-name>/.meshive-connect instead of the space outside the volume. The folder is created when you log in, counts toward the volume’s size, and is safe to delete. Pods that mount the volume see it too, but nothing in a pod uses it.

You might see	It means	Do
Connection restarted — out of memory	Something in the channel used more than 2 GiB of memory	Run fewer downloads at once; do heavy processing in a pod
Connection restarted — more than 1 GiB written outside the volume	A download or install filled the space outside /mnt/<volume-name> — often a download into the home folder	Download into the volume, as above
Your SSH session dropped and installed tools are gone	The channel restarted	Reconnect and reinstall the tools — they live outside the volume

Choosing between the two

• Local — training scratch space, model weights next to the GPU, anything latency-sensitive. Remember it pins the pod to that machine: to move GPUs, copy the data out first (why).
• Network — datasets shared across pods, data that must outlive any single machine, spot workloads (local volumes aren’t offered on spot — they’d be lost on reclaim).