AI and your data
Last updated: 28 October 2026
1. Overview
Instama B.V. ("Instama") uses AI models to create and edit marketing content. This page lists the models the platform uses, how each one is reached, and what the model providers may do with the data they receive. We describe only what is in place today.
Every AI call starts on Instama's servers. Model provider keys never reach the browser, and a model can act only through the same permission-checked tools as a person: it never gets direct access to the database.
All Client Data is stored in the European Union. To fulfil a request, the platform sends the content that request needs to a model provider, which may process it outside the European Union under the safeguards described in our Privacy Policy.
2. Models and what happens to your data
The models the platform may use, who makes them, what we use them for, and what their provider may keep. Instama does not use Client Data to train AI models, and no provider below may train on it. Image and video models are used only once the service that runs them has bound their maker to no training and to its security and data-protection obligations; we check this before each release. The list changes as models improve; this page is updated when it does.
| Provider | Model | Used for | Runs in | Remarks |
|---|---|---|---|---|
| Anthropic | Claude | Assistant, writing, planning, review | The provider’s regions; an organisation can keep it in the EU, the United States, Australia or Japan | No training. Kept up to 30 days for abuse monitoring. |
| OpenAI | GPT | Writing, review | The provider’s regions; an organisation can keep it in the EU or the United States | No training. Kept up to 30 days for abuse monitoring. |
| Gemini | Captions, image understanding, translation | The provider’s regions; an organisation can keep it in the EU or the United States | No training. Kept up to 30 days for abuse monitoring. | |
| Cohere | Embed, Rerank | Search: embeddings and ranking | The provider’s regions | No training. Kept up to 30 days under the provider’s defaults. |
| Perplexity | Sonar | Web search and reading web pages on request | United States | No training. Kept under the provider’s defaults. |
| TypeSafe AI | Jev | Automated decisions inside the platform | United States | No training. Zero data retention: nothing is kept beyond the request. |
| Gemini 2.5 Flash Image | Image generation and editing for an organisation that keeps its images in one region | The provider’s regions; an organisation can keep it in the EU or the United States | No training. Kept up to 30 days for abuse monitoring. | |
| Nano Banana | Image generation and editing | United States | No training. Kept only as long as needed to make the result; the files we send expire within 24 hours. | |
| OpenAI | GPT Image | Image editing | United States | No training. Kept only as long as needed to make the result; the files we send expire within 24 hours. |
| Kuaishou | Kling | Video generation | United States | No training. Kept only as long as needed to make the result; the files we send expire within 24 hours. |
| ByteDance | Seedance | Video generation | United States | No training. Kept only as long as needed to make the result; the files we send expire within 24 hours. |
| Luma | Ray, Photon | Reframing video and images | United States | No training. Kept only as long as needed to make the result; the files we send expire within 24 hours. |
| fal.ai | Reframe | Reframing images | United States | No training. Kept only as long as needed to make the result; the files we send expire within 24 hours. |
| Ideogram | Ideogram | Separating the text of an imported image into layers | United States | No training. Kept only as long as needed to make the result; the files we send expire within 24 hours. |
| Alibaba | Qwen Image | Splitting imported artwork into layers | United States | No training. Kept only as long as needed to make the result; the files we send expire within 24 hours. |
| Pixelcut | Pixelcut | Background removal and product photos | United States | No training. Kept only as long as needed to make the result; the files we send expire within 24 hours. |
| Bria | Bria | Video background removal | United States | No training. Kept only as long as needed to make the result; the files we send expire within 24 hours. |
| Topaz | Topaz Upscale | Upscaling | United States | No training. Kept only as long as needed to make the result; the files we send expire within 24 hours. |
| OpenAI | Whisper | Speech to text | United States | No training. Nothing is kept. |
| Firecrawl | Web reader | Reading a brand’s public website during onboarding | United States | Only public web pages are sent; the provider may cache them. |
3. Principles
Transparency
We publish which providers the platform uses and how they are reached, and we describe what is not yet in place as plainly as what is.
Purpose limitation
Client Data is processed only to provide the platform. A model provider receives only the content that the request in question needs.
Data protection
Client Data is stored in the European Union. Processing follows the commitments in our Terms and Conditions and Privacy Policy.
Oversight
Before a model is added, we benchmark it against the models it would replace and check the terms under which its provider processes data.
Organisation control
Organisations decide who may use AI features through roles and brand membership. An organisation can also choose which models it uses, per kind of work, and keep its text models in one region (the EU, for instance): a model that cannot run there is then not used, and a request that cannot be served there fails rather than leaving the region.
4. Changes
We update this page when the inventory or our practices change. Material changes that affect how Client Data is processed are announced by e-mail or in the platform before they take effect.