Responsible AI
Custom models are built from your data, so trust is central to our work. These are the commitments we make to every client, written into our agreements.
Your data and your model belong to you
Under our standard contract, you retain ownership of all data you provide, and you own the model weights we produce for you. The only limits on how you use those weights come from the license of the open-weight base model, which we review with you before work starts.
We do not claim rights to reuse your weights, your datasets, or your evaluation sets.

No training on your data for anyone else
Your data is used only for your project. We never combine it with other clients' data, never use it to improve models delivered to other clients, and never sell or share it. Project environments are kept separate for each client.
Data deletion on request
At the end of an engagement, we delete your data and intermediate artifacts from our systems unless you ask us to keep them for ongoing support. You can request deletion at any time by writing to privacy@getomnilo.org. We confirm deletion in writing. Backups are purged on their normal rotation schedule, which does not exceed 30 days.
Security practices
- Encrypted data transfer and encrypted storage.
- Access limited to the engineers assigned to your project.
- The option to run all work inside your own cloud account.
- Personal information removed or masked from training data where required.
Use cases we decline
We will not build or adapt models intended for:
- Weapons development or planning violence.
- Surveillance, tracking, or profiling of individuals without their knowledge.
- Generating deceptive content, including impersonation, fake reviews, or disinformation.
- Any illegal activity, harassment, or discrimination.
If a request falls into a gray area, we discuss it openly and may decline the work.
Respecting model licenses
Open-weight models come with licenses that set rules on commercial use, attribution, and acceptable use. We check these licenses for every base model and teacher model, explain the relevant terms to you in plain English, and only use models whose licenses permit your intended use. Where a license requires attribution or passing on usage restrictions, we include that in your deliverables.
Honest evaluation
Every model has limitations. Our evaluation reports include failure cases, not only averages, so you can decide where the model should and should not be relied upon. We recommend human review for high-stakes decisions.
Questions about how we handle data?
We are happy to walk your security or legal team through our practices before any data is shared.
Book a free scoping call