Sign inRequest access
Use case · For privacy

Use a cloud AI without handing it your private details

The best cloud models write beautifully, and you’d still rather they didn’t see your Social Security number. Turn on Sensitive information checks, and before a message or memory goes to a public AI provider, Geyser looks for private details in it. You choose what happens when it finds some. With “Use a private model for the whole request”, that request goes to your own model, and everything else still goes to the cloud model you pay for.

Request accessMore use cases

“I write with Claude most of the time. When the checks find private details, I work with the model on your Mac.”

Wren

Wren wrote the appeal to the insurer in minutes. The message with your Social Security number in it was read by the model on your Mac, and Claude never saw it.

A week of paperwork with Wren
  1. Mon 8:10 pm
    YouChat with Wren

    Asks Wren to appeal a denied insurance claim, and types in the claim details, Social Security number included.

  2. Mon 8:11 pm
    WrenA model on your Mac

    The checks find your Social Security number in the message, so the whole request goes to the model on your Mac, as you chose.

  3. Mon 8:14 pm
    WrenChat with Wren

    Drafts the appeal and quotes the line of the policy that covers the claim.

  4. Tue 12:30 pm
    WrenClaude

    You ask for ideas for a birthday dinner. The checks find nothing, so Wren uses Claude.

  5. Tue 6:02 pm
    WrenNeeds you

    “The insurer hasn’t answered. Want me to draft a follow-up that mentions the date you first filed the claim?”

How it works

The pieces behind the story

Where it runs

Wren uses Claude for most work and a model on your own Mac for anything the checks flag.

Who does it

Geyser runs the checks itself before anything goes to a public AI provider, so it doesn’t depend on Wren remembering to.

What you control

You choose what happens when the checks find something, from “Record the finding and allow sending” to “Block sending to public AI”.

What the checks find goes to a model you run yourself, with no per-token bill, and the rest goes to the AI plan you already pay for.

Set it up

Getting set up

  1. 1

    Switch on “Run local models here” on your Mac and choose a model.

  2. 2

    Turn on Sensitive information checks in your workspace’s AI settings.

  3. 3

    Choose “Use a private model for the whole request”.

Keep exploring

More use cases

Browse all use cases

AI for files you’d never paste into a chatbot

Ada reads your tax papers and medical bills with a model on your own Mac, so none of it goes to an outside AI company.

Read the story

Ask an Agent to build the tool you need

Tell an Agent about the tracker you wish you had. Pia builds it in the conversation, and it reads your receipts with the model on your own Mac.

Read the story

Follow every coding session from your phone

Claude Code on your laptop and Codex on the build box keep going while you’re out. Your Agents tell your phone which session finished and which one is waiting on you.

Read the story
Make a start

Start with one Agent and one real job

Set up a workspace, give an Agent something useful to do, and see how far it gets.

Request accessSee pricing