Privacy and data
Your Models are yours. Facet uses what you put in only to build, check and serve your Models, never to improve anything for anyone else. The full terms are in the privacy policy; this page is the short version for builders.
What Facet keeps
| What | Kept | For how long |
|---|---|---|
| Your Model: description, Instructions, evals, examples, versions | Yes | While the Model exists; unused versions move to cold storage, so you can roll back |
| Calls to a Model: the messages and the answer | Not stored. They pass through to answer the call; Modal, which runs it, keeps nothing | Not kept, unless data capture is on |
| Calls to a Model: version, key, token counts, latency, request ID | Yes, to bill you and show usage | 90 days |
| Examples sent to the open models that write and check examples | No: only hosts that keep nothing once they answer | Not kept |
Who sees it
-
You and your workspace, by role: Owners, members and viewers.
-
Facet’s agent, which plans and makes changes to your Model. It never sees hidden examples.
-
Facet’s checks, which are the only thing that reads hidden examples.
-
The services that run Facet, each only for its part. Once you say yes to a draft, your rows go to:
- OpenRouter, which sends every model request, only to hosts that keep nothing once they answer: the open-weight models that write training examples and check answers, and Claude, which runs Facet’s agent on Amazon Bedrock or Google Vertex. The agent never sees hidden examples.
- Modal, the GPUs that train, check and serve your Model.
- Cloudflare R2 and Neon, where files and records are stored, in the US.
Google Cloud runs the app, WorkOS signs you in, Stripe takes payments and Resend sends email. The privacy policy lists what each one sees.
Hidden examples
When you add examples, some are hidden. Once hidden, an example is read only by Facet’s checks: not by you, not by Facet’s agent, and never used to build a version. That’s what makes a green eval honest: no version can learn its own test. Exports leave hidden examples out for the same reason.
Keys
API keys start with fct_. Facet stores only a hash of each key, so a lost key can’t be shown again; make a new one and revoke the old one under Settings, Keys. A key can be limited to one Model, and a viewer key only reads.
Delete
- A Model: archive it from its Settings. To delete it for good, a workspace Owner shows Archived in Models, opens the Model’s menu and picks Delete for good, typing its name. You have 7 days to undo it with Keep it; then its Versions, evals, examples, training data and history are deleted, and backups expire within 7 more days. Billing records keep what it cost.
- A workspace or your account: delete it from Settings. You have 7 days to undo it; then its data is deleted, and backups expire within 7 more days.
- A copy of your data: export any Model (see Export your model), or write to privacy@halfspacelabs.com.
Next steps
- Privacy policy: the full terms
- Add examples: what to leave out of uploads