Privacy Policy
- Version:
- 2026-09-08
- Status:
- Current
- Published:
- 2026-10-02 06:33:59.909Z
- Earliest permitted effect:
- 2026-10-02 06:33:59.909Z
- Actually activated:
- 2026-10-02 06:33:59.909Z
- SHA-256:
- d697bb00f67c167ed7714da0d6b6019aa7697f603e19a4dae68b8826fa0a0e89
Publication and an earliest-effect date do not mean activation. This version has an immutable document link.
What this covers
UltimateModel AI lets you fine-tune and serve your own AI model on your own data. This policy explains what we collect, where it goes, who else processes it, and how to get it deleted.
Two distinct groups of people are covered. You, our customer, and your end users — the people who talk to a model you have deployed, including through our embeddable chat widget. The section on captured traffic is the important one for them, and it is your responsibility to disclose it to them.
Data you give us directly
Account data — email address, name, hashed password. Passwords are stored as bcrypt hashes; we cannot read them.
Training data — documents, pasted text, URLs, voice-dictation transcripts, and supported audio/video files you explicitly choose to upload, plus derived text chunks, reviewed training/evaluation records, and adapter artifacts. Browser dictation sends its transcript rather than its raw audio. File transcription and scanned-document OCR use locally configured processors and model assets, without an automatic external-provider fallback.
Payment data — handled entirely by Stripe. We store a Stripe customer identifier and your plan. We never see or store card numbers.
Contact and support requests — the name, email, company, message and request reference you submit for follow-up.
Conversations — ordinary chat and widget messages retained for conversation history and project analytics. This is separate from the disabled automatic training and scoring capture described below.
Where your data is stored
Each customer gets a dedicated, access-isolated Google Cloud Storage bucket. Your training data and model weights are not commingled with any other customer’s. Storage and transport protections follow the deployed Google Cloud configuration rather than a hardcoded cipher claim. The reference deployment uses us-central1 for compute and Cloud SQL, and US multi-region customer buckets. These are not the same location guarantee. Actual residency, backups and external export destinations require deployment and contract verification.
Your data is never used to train another customer’s model, and never used to train a general-purpose model of ours.
Production traffic and evaluation evidence
Automatic provider scoring and raw inference traffic sampling are not active. The current production path does not persist raw prompts and responses for automatic scoring or training. The historical capture flag is contained until a bounded, sampled, privacy-reviewed manifest is implemented.
Imported evaluation evidence must be project-scoped, minimized, and redacted. Only salted example identifiers, scores, provenance, calibration, and safety results are persisted by the promotion evidence workflow.
If your end users are members of the public — for example through the embeddable widget — you are the controller of their data and you are responsible for telling them how the live inference request itself is processed, that ordinary conversation text is retained, and how your users can exercise their rights, including any additional retention you introduce.
Subprocessors
We share data with the following processors, and only for these purposes:
Your rights and how to use them
You can access, correct, export or delete your data. Model weights and training artifacts can be exported when a completed export artifact exists. Account settings can create an erasure request, immediately disable access, and issue a private receipt for progress. For questions or requests that cannot be initiated in the app, submit a privacy inquiry (/contact?inquiry=privacy) or email privacy@ultimatemodel.ai . Account erasure is asynchronous: provider revocation, billing cancellation, object deletion, active database purge, and restore suppression are each verified before the receipt says completed.
Captured traffic already stored under a project is covered by erasure: project deletion removes its captured logs, and account deletion removes captured logs across the account. Some financial records at the payment processor and minimal security or audit facts may be retained under the documented policy; the deletion receipt identifies each retention exception and its expiry instead of reporting it as deleted. This operational description still requires legal review and is not a claim of compliance with any particular law.
Retention
Active customer content remains while the account is active. Once erasure begins, access is frozen and active-system content is removed only after each external and local deletion is confirmed. Google Cloud SQL is configured for seven days of point-in-time recovery and fourteen retained daily backups. A subject suppression remains for thirty days after completion so a restore can reapply the deletion before traffic resumes; application code cannot selectively erase an immutable historical backup.
Versioned platform buckets retain at most five archived generations. Tenant buckets are not versioned and the erasure worker verifies their absence. Minimal redacted security/audit facts are configured for up to 365 days. External financial records may be retained for up to seven years pending the applicable financial-record policy and legal review. Customer prompts, responses, tokens, and files are not retained as financial facts.
Contact
Questions about this policy: submit a privacy inquiry (/contact?inquiry=privacy) or email privacy@ultimatemodel.ai . See also our Terms of Service (/terms).
Explicit sharing and request records
Projects are private by default. An owner can explicitly share selected project configuration, source metadata, and bounded evaluation summaries with eligible workspace members. Membership alone grants no project access. Editors can prepare data for owner acceptance; this does not grant owner credentials, original artifact exports, paid training, or serving control.
Support, rights, evidence-review and commercial requests include customer messages, private operator notes, assignment, scheduling, quote acceptance, policy assent, and notification delivery receipts. These records support request follow-through and are included in the applicable access and erasure workflows. Private email receipt links permit only the named request history and replies; they cannot accept a commercial quote or access the account.