Assets

Workspace › Assets is where your datasets, model weights, LoRAs, configs, and job outputs live. One place, one list, shared by everyone in the workspace — the same asset can feed a Pod, a task, or a download.

The three layers

Layer	What it is	What you see
Asset	A named series — “cat-lora”, “training-set-v2”	One row in the list
Version	One immutable snapshot of that series (v1, v2, …)	The version stack in the detail drawer
File	One file inside a version, with its relative path	The file list under a version

Uploading a folder produces one version with many files, not many assets — directory checkpoint formats stay intact.

Versions are never edited in place. New content is always a new version, and old versions stay downloadable until you delete them.

How new versions are decided

• Same source, same name → vN+1. A Pod that harvests checkpoints/ on every run keeps appending to the same series.
• Different source, same name → a new asset plus a merge prompt. Meshive does not silently mix two origins into one series. The list shows a banner with two choices:
  • Merge as vN — move the new content into the existing series (all versions are renumbered in order).
  • Keep separate — leave the two series independent and dismiss the prompt.
• Renaming is free and does not rewrite history. Because the name is what identifies a series, a Pod that keeps writing the old directory name will start a fresh series after a rename — that’s intentional.

Types

Each asset carries a type — dataset, model, adapter (LoRA), checkpoint, output, config, or file. It is a hint, not a rule: it drives filtering and the default mount location when you attach the asset. Meshive guesses from the file extensions (or from the template’s declaration for harvested output), and you can correct it in the detail drawer.

Badges

Badge	Meaning	What to do
pickle · runs code on load	The file is a Python pickle format (.pkl, .joblib, .pth, some .bin) — loading it executes code. detected means we read the file header and confirmed it; suspected means we inferred it from the name	Only load assets you trust. Prefer safetensors when the publisher offers it
format not verified	The file has no extension, so we can’t say what it is	Neutral — common in directory checkpoint trees
in use	A running Pod or a non-terminal task is consuming this asset	Delete is blocked until it’s released
frozen	The asset is under abuse review (Terms §10)	Everything except metadata edits is locked
M / S3	Stored in Meshive-managed storage / registered in your own bucket	Only M bytes are billed as asset storage

Where the bytes live

Storage	Created by	Billed	Deleted by Meshive
Managed	Web upload, Pod harvest, task output, imports, subset fetches	Yes — per GB·month (details)	Yes, when you delete it or when the unpaid-balance lifecycle runs
Your bucket	In-place registration and prefix references (details)	No — you already pay your provider	Never. We only read

Deleting

Deleting an asset removes it from the list and stops storage charges immediately. Managed bytes are then physically removed by a background cleanup pass; objects in your own bucket are left untouched.

• Delete is blocked (409) while a Pod or task is using the asset. Stop or finish the workload first.
• Individual versions can be cleaned up only while they are failed or incomplete uploads — completed versions are removed by deleting the asset.
• Deletion is not recoverable. Download or export anything you want to keep.

Where to go next

• Getting data in — upload, harvest from a Pod, import from Hugging Face or CivitAI, register or reference data in your own bucket.
• Using assets — attach to a Pod, inject into a running Pod, feed a task, download or export.
• Billing, limits & safety — what storage costs, what happens when credit runs out, and the abuse-report flow.