Your data
You decide where your data goes.
Every assistant in Flow has its own data policy. Every model is labelled with where it runs, who owns the provider and what it keeps. You choose. You don't have to hope.
01 · Home region
One home region. Set once.
When your company signs up, you choose where your data lives: the EU (Frankfurt) or the US. It can't be changed afterwards, and your files, index and memory are stored in that region.
02 · A data policy for each assistant
Where can this assistant's data be processed?
That's the question Flow asks when you set up an assistant. Three answers:
Where can this assistant's data be processed?
Details: region · owner · retention · price → Show
Within the policy, you choose how thorough each task should be: Quick, Standard or Thorough. That decides which class of model is used and what it costs. Model names stay out of the way until you ask for details.
03 · Said plainly
What our words mean.
- «In the EU»
- means processed in the EU. Some of those providers are American and subject to US law, whatever the region. We label them, so you can choose.
- «Our own model»
- means we send tasks to your endpoint. Running everything on your own hardware comes with Nrth Node in a later version.
- «No training»
- means no model is trained on your data. Flow learns from the knowledge you connect, the memory you can read and the skills you approve.
04 · Every model, labelled
Every model, labelled.
| Provider | Processed in | Provider based in | Available under |
|---|---|---|---|
| Claude via AWS Bedrock | EU (Frankfurt) | United States |
|
| OpenAI with EU data residency | EU | United States |
|
| Mistral | EU | France |
|
| Open models on Scaleway | EU | France |
|
| Open models on OVHcloud | EU | France |
|
| Your own model | Your endpoint | You decide |
|
05 · How it learns
Three ways it learns. None of them is training.
Knowledge.
The sources you connect: uploaded files, Google Drive folders, SharePoint libraries. You see each source, when it was last synced, and any errors. Remove a source and its index is deleted.
Memory.
What you've told it and the feedback you've given. Two layers: about the company, and about each assistant. Every line can be read, changed or deleted.
Skills.
Tasks that work, saved as reusable skills after you approve them. Each shows when it was last tested.
06 · What it never keeps
What it never keeps.
Some things are never written to memory, whatever you tell it:
- ID numbers
- Payment card and account numbers
- Health information
- Information about children
- Special categories of personal data under GDPR article 9
07 · How it's built
How Flow is built.
Flow is being built to these rules:
- Every task runs in its own isolated environment, deleted when it's done.
- Your files, knowledge index and memory are stored in your home region.
- Keys to your own model and connected accounts are kept in an encrypted vault, never in the task.
- Hard spending limits per company and per task.
- Actions outside Flow, like sending on your behalf, need your approval.
- Every skill is tested against attempts to slip instructions in through documents and web pages.
A full security page and a data processing agreement (DPA) will be published before launch.
Pilot program
Try Flow free, together with us.
We're onboarding the first companies in the Nordics and the US before launch in May 2027. Free during the pilot, 50% off the first year after launch, and a direct line to the people building it.