Control which providers see your data
Found this helpful? Share it:
Found this helpful? Share it:
Different AI providers treat your prompts differently: some keep nothing, some retain them briefly, one or two may train on them. idapt labels every provider from its own published terms, quotes the sentence behind each label, and lets you restrict routing to the ones that meet your bar 🛡️
Provider data policies show up throughout the product, so you can check before you chat:
Model cards in the model catalogshow a No-train badge when the model's best route qualifies.
Model detail pages list every provider serving the model; open a provider row to see whether it trains, how long it retains data, and a link to its published policy.
The provider directoryand the privacy pageshow the full labeled table.
No-train means the provider doesn't use your data to train models, but may retain it briefly (often ~30 days) for abuse monitoring.
Instead of checking labels model by model, set a floor once and let idapt enforce it on every request:
Open your workspace's AI Routing
Find the Data policies card
Turn on Only endpoints that don't train on my data
The floor applies to every modality: chat, agents, the API, and image, video, speech and transcription too. Providers whose policy we could not verify are excluded, never assumed safe. The line under the toggles shows how many models the floor leaves you, and the preview shows exactly which providers remain.
Floors only ever tighten: an agent can be stricter than its workspace, never looser. Models running on your own computer always qualify.
A provider marked Policy not published is not an accusation. It means we read its terms and found no statement we could quote, so we will not route your data there while you are asking for a guarantee. May train is different: that is a provider saying so in its own words.
You can also require that a provider be legally established in specific countries. Open AI Routing, find Only providers headquartered in, and pick the countries you accept.
This filters on where a company is established, which is a checkable fact. It is not a data-residency guarantee: most providers do not publish their datacenter regions, and we cannot verify where an individual request ran. Where a provider does publish its regions, we show them on its row so you can see them.
Don't want a particular provider handling your data at all? Block it in AI Routing preferences and idapt never routes to it, for any model.
You can also block a provider for a single model from the providers list on that model's details page.
The strongest control is not routing to a provider at all. Private AI runs open models on a computer you own, so no third-party AI provider ever receives the prompt. See Run models locally.
For maximum privacy, run open models on your own computer, where no third-party AI provider is involved at all. See Run models locally.
Related articles
Your data and privacy
What idapt stores, who can see it, how AI providers handle it, and the controls that keep you in charge.
Choosing models
Browse 200+ AI models, read their cost bands, set the speed dial, and pick the right model for each conversation.
Run models locally
Install open models on your own computer with Ollama and run them privately in idapt: free, with no rate limits.
Was this helpful?