The intelligence you build stays with you.
Philotic models run on open weights and our own inference code, so they can run on our dedicated hosting, in your cloud, or on your servers. Customer builds call no third-party model APIs.
A model trained on your operation learns how your people decide. That should never leave your control.
Four kinds of control
What Philotic commits to, in writing.
Data
Your data goes where you say, and trains only your models.
- Nothing from your operation trains or evaluates a general Philotic model or another customer’s model. That covers every signal: photos, decisions, edits, and the accuracy metrics computed from them.
- Data a third party controls, such as a partner’s or a regulator’s, stays out of analysis and training until you confirm you have the right to use it. That rule is enforced on the server, not only in policy.
- We don’t publish results specific to you without your written approval.
Models
You own what’s trained on your data, on a foundation you chose.
- Adapters trained on your data are yours. An adapter is a separate weight file from the model beneath it, so ownership follows the files.
- You choose the open foundation. We test the candidates on your task and share the results before you decide.
Compute
You choose where models run and who operates them.
- Philotic-hosted on capacity dedicated to you in US regions, in your own cloud account, or on your own servers.
- Customer builds call no third-party model APIs. Inference runs on open weights and our inference code, inside the boundary you set.
Operations
Nothing changes in production without your approval, and what you own leaves with you.
- AI stays off until a model passes an accuracy bar you approve, on held-out examples. New versions go live through the same gate.
- During the term you license our domain models with every update, and the inference code to run them in your own cloud or on your own servers.
- If you leave, you keep your data, labels, evaluation sets and adapters. Retrain on a foundation model of your choice, or buy out the version you are running.
Three ways to deploy.
The same model and the same approval gate in each. What changes is who operates the hardware.
Philotic-hosted, dedicated
Your cloud
Your servers
What you keep, and when.
Sovereignty is only real if you can keep going when a provider can’t. Here is what you hold during the term, and what you keep if you leave.
- Your data and adaptersYour data, labels, labeled sets, evaluation sets, configuration and the adapters trained on them. Owned by you, always.
- Domain modelsThe Philotic models your adapters sit on. Licensed to you for the term, with every update.
- Inference codeThe code and container images that serve your models. Licensed for the term, with deployment runbooks delivered to you.
- If you leaveKeep everything that is yours and retrain on a foundation model of your choice, or buy out the frozen version you are running. No free license continues.
- A bought-out versionInternal use only, not transferable, no updates, and no copying or distilling the model. It runs without our platform.
- Never includedThe platform, the training pipeline and our infrastructure stay ours.
Put your models where your data lives.
Tell us where your data sits today and what your security team requires. We’ll show you how the deployment fits inside it.