Generated vs Saved Artifacts
There are two ways content becomes an artifact: generated (the model produced it and the platform auto-suggested saving it) or user-saved (you explicitly clicked Save). Knowing the difference helps you organize your vault.
User-saved artifacts
You hit the Save button on a chat reply, or used + New Artifact directly. You picked the title, type, and tags. These are explicit, intentional, and tend to be what you actually want long-term.
Characteristics:
- Title set by you
- Type confirmed by you (auto-detection is just a default)
- Tags reflect your organization scheme
- Usually higher signal — you didn't save it on accident
Generated artifacts
Some chassis (notably Atelier and Studio) auto-suggest saving certain model outputs:
- Generated images go to the artifact vault automatically
- Long code blocks may prompt with "Save as artifact?"
- TTS audio gets saved with playback metadata
Characteristics:
- Title auto-generated from prompt or content
- Type set automatically
- No tags by default (you can add them)
- Mixed signal — useful but easy to accumulate clutter
Why distinguish them
Three reasons:
- Cleanup. Generated artifacts pile up faster. Periodic reviews of your vault are easier if you can filter "show me only what I explicitly saved."
- Searching. User-saved artifacts tend to have higher-quality titles and tags, so they rank better in search results.
- Sharing intent. When export/sharing rolls out, you'll want to share user-saved (curated) work, not generated (raw output).
Currently the vault doesn't visually distinguish between the two, but the source field is in the data model and a filter is on the roadmap.
When generation auto-suggests saving
| Content type | Auto-suggests save? |
|---|---|
| Plain text reply | No |
| Code block (any size) | Most chassis: only on copy |
| Long markdown reply with structure | No (use the explicit Save button) |
| Generated image | Yes, in Atelier; on-demand elsewhere |
| Generated audio (TTS) | Yes, in Studio; on-demand elsewhere |
| JSON output that looks like configuration | No |
The platform errs toward NOT auto-saving so your vault stays clean. If you want something saved, you save it.
Override: don't auto-save
In chassis where some content auto-saves (Atelier, Studio):
- Settings → that chassis's preferences may have a toggle
- Otherwise, you can delete after the fact
Override: always save
If you want the platform to auto-save more aggressively:
- Most chassis don't expose a "save everything" toggle
- The workaround: build the habit of clicking Save on every reply you might want later
Best practice for organization
- Tag generated artifacts as soon as you save them so they're searchable later
- Periodically prune — open vault, sort by oldest, delete what's no longer useful
- Pin user-saved artifacts you reference often — the visual prominence helps
- Use distinct tags for generated vs handpicked content (e.g.
genfor generated,keepersfor handpicked)
A note on artifacts as recall context
Both generated and user-saved artifacts are equally eligible to be surfaced as recall context during chat. The recall layer doesn't know which was which — it just looks at semantic relevance. If a generated artifact is highly relevant to your current message, it surfaces.
This means: prune generated artifacts that are bad signal. Otherwise the model might pull in a bad-quality artifact as recall context.
Where to next
- What Are Artifacts? — concept overview
- Vault Overview — UI walkthrough
- Using Artifacts in Chat — recall + injection
- Search and Tags — organizing for findability