Privacy & control

Your AI should answer to you, and nobody else

Most AI assistants start by asking for everything: your inbox, your files, your bank, and a copy of it all on their servers. Geyser starts from the other end. Your Agents work on your own computers, you choose which AI sees which work (including models that never leave your hardware), and the rules are enforced by Geyser rather than left to the AI. We never sell your data, and our own support team has to ask your permission before looking.

“I keep what I learn about your business in the place you chose for it, and I can’t change my own rules.”

June
Explore an example setup
Your workspace
Company knowledge

Stored by Geyser

Agent computers

Your own computers with Geyser Host

AI & connected apps

AI services & apps you approve

Geyser service operations

Accounts, billing, connections, software updates, and service health.

Geyser stores your workspace’s data, and your Agents work on your own computers with Geyser Host. You choose which AI services and apps they use.

Choose where your data is stored

Your conversations, documents, and what your Agents learn are kept in their own storage, which we call a Customer Cell. We can host it for you in our US region, or you can keep it on systems you control, including your own Mac or Linux computer.

Shared knowledge, within permissions

People and Agents can find the information they’re permitted to use. A private conversation stays private when it’s linked to related information, and asking another Agent for help doesn’t give access to that Agent’s memory.

Choose which services handle your information

AI models, voice services, search, and connected apps each need some of your information to do their part of a task. You can use services you’ve approved, or supported options running on your own systems.

Check before anything goes to a public AI

Turn on Sensitive information checks and Geyser checks for private details before a message, memory, or file reaches a public AI provider. You decide what happens when it finds some: record it and send, remove what the built-in checks match, use a private model for the whole request, or block sending to public AI.

Decide what your Agents can do

Geyser checks permissions itself, separately from the AI, the same way for every AI engine, so an Agent can’t argue its way past them. You choose which actions wait for a person’s approval, and an Agent that can see a screen still needs separate permission to control it.

Watching a screen isn’t the same as driving it

Only the people allowed to control a computer get its mouse, keyboard, and file transfers. Everyone else you let in sees a view-only picture, and Geyser support can only watch, with a grant your workspace Owner approves.

Agents don’t take orders from strangers

Email, texts, and meeting speech are treated as untrusted. When someone outside your team starts a conversation, the Agent doesn’t take their word for anything: actions wait for your approval. No Agent can change its own rules.

Your keys stay out of the Agent’s reach

API keys you add to your workspace are stored encrypted by Geyser and never copied onto an Agent’s computer, so an Agent can use your AI account without holding the key.

Spending limits an Agent can’t talk its way past

Each Agent has a daily model allowance and the workspace has spending limits. Any Admin can pause every Agent at once, and only the workspace Owner can switch them back on.

Visibility without a back door

Our support team asks first

To run the service, Geyser needs your account, routing, billing, and health information. Your content is kept apart, in its own home with its own access controls.

If support needs to open your content, your organization’s owner approves it first, and we keep a record of that access.

See how the boundary works

What a support request has to spell out

Who
A named operator
Why
A defined support purpose
What
The information or tools you approve
How long
A set end time
Decision
Your organization’s owner
Good questions

A few useful answers

If an Agent runs on my computer, does everything stay there?
It can. The Agent does its work on your computer. Keep your data home on your own hardware too, use private models, and the work stays in the building.
Can our data stay in the EU?
Yes, on your own hardware. Keep your workspace’s data home on a computer in your own country, and your Agents work on your own computers in your office. Pair them with Mistral set to EU regional inference, or a model you run yourself, and your files, your Agents’ work, and the AI reading it all stay in Europe. Account, billing, and phone service records are handled by our central service as the Privacy Policy describes.
Can personal details stay away from AI companies?
Yes. Give that work to a model on your own hardware, or turn on Sensitive information checks so what they find goes to your own model, or doesn’t go out at all. An Agent can build an app that reads ID scans, pay stubs, or patient notes with your own model and passes the Agent only the summary the job needs.
Is Geyser HIPAA compliant?
Yes. A clinic or practice can use Geyser with patient information, and you can keep the most sensitive records on your own hardware with a private model.
What happens when I use an outside model or app?
The service gets only what the task needs, through the route you approved. When something shouldn’t leave your hardware, give that work to a private model instead.
What does Geyser keep centrally?
The shared service holds the operational information it runs on: identity, directory, billing, routing, releases, and health. Approved provider bridges may also process content as it passes through. The technical architecture covers storage and processing separately.
What happens to voice and screen observations?
The voice path processes audio transiently. The text it produces stays with the conversation and follows that conversation’s retention rules. Screen observations are captures limited to a set scope, and the observing Agent’s model can process them. You pick the speech route and the reasoning model separately.
Can I use my own data to train a model?
Yes, through a dedicated model-training workflow that you choose to run, with data you have permission to use. It has its own steps for examples, review, training, and evaluation. For how the service handles information in general, see the Privacy Policy.
Can I move or delete our work?
Yes. Export your workspace whenever you like, or move your data home between our cloud and your own hardware. Our team can walk you through deleting it, too.
What is Private Cell Strict?
It’s a supported deployment setup with default-deny egress, inference on systems you control, and restrictions on shared relays, provider bridges, and content federation. Geyser still runs the limited central operations the service needs.
Make a start

Draw the lines yourself

Tell us where the data should live, which AI may see it, and which systems your Agents need to reach.