Skip to content
Desktop bridge 1.5.8Human-approved execution

Cloud AI for work that has to touch your computer.

Run AI tasks in the browser, then approve the exact file reads, writes, and commands before anything happens on your machine.

macOS, Windows, Linux and Android·No API keys·Every local action asks first
/app/tasks/9f2cCloud reasoning
Turn the Q3 CSVs in ~/reports into a single PDF summary.
ROUTED → CLOUD REASONING · FAIR USE OK
I found 3 files. I need to read them on your computer and run one command to build the PDF. Approve on your desktop app to continue.
Desktop bridge · studio-mbpLinked
Approval requested
pandoc q3-summary.md -o q3-summary.pdf

Reads 3 files. Writes 1 file. No network access.

Approve onceAlways allowDeny
Audit trail
  1. read ~/reports/q3-emea.csv
    2.1 MB · approved by you 14:02
  2. read ~/reports/q3-amer.csv
    1.8 MB · approved by you 14:02
  3. run pandoc q3-summary.md -o q3-summary.pdf
    awaiting your approval
  4. write ~/reports/q3-summary.pdf
    blocked until the command is approved

No personal API keys

Model access is included in your subscription. You never paste a provider key.

Local work needs approval

Every file read and every command is shown in full and waits for you.

No surprise token bills

One subscription price. Heavy use is slowed under fair use, never billed extra.

Clear audit trail

Every task keeps a timestamped record of what ran, where, and who approved it.

What you can run

The agent shows its boundary before it works.

Cloud planning, local approval, and a readable trail for every handoff — this is what that looks like in practice.

01 Plan02 Ask03 Run
Files

Local file summaries

Review project folders, compare reports, and generate clean notes after you approve the exact files to read.

~/clients/acme/*.pdf
Commands

Command-assisted tasks

Let the agent prepare scripts, run checks, convert files, or build artifacts only after the command is shown.

npm test -- --reporter
Audit

Tracked handoffs

See what the agent requested, what you approved, what changed, and where the output landed.

ApprovedWaitingBlocked
Create

Images, too — priced per image, never per token.

Pick a model, write the prompt, and see the price before anything runs. Images are paid from prepaid credit you top up yourself; nothing is billed automatically.

See what you can make
How it works

From signup to a finished task without losing control

The desktop bridge is a small app. It links your computer to your account, then stays idle until a task needs local access and you approve it.

  1. Create your account

    Choose a plan with model access included.

  2. Install the bridge

    Use the build for macOS, Windows or Linux.

  3. Link your computer

    Confirm the short code in your browser after checking the device details.

  4. Run approved work

    Prompt in the browser. Approve only the file reads, writes and commands you expect.

Cloud reasoning, local execution

Models run in the cloud. Files and commands run on your machine, under your account, behind an approval prompt.

Model access
Included
Local execution
Per approval
Provider keys
Not required
Audit trail
Automatic
Questions

The things people ask first

Do I need my own API key?

No. Model access is included in every plan. You never paste a provider key.

Does the agent run things without asking?

No. Reading a file, writing a file and running a command each require an explicit approval in the desktop app. Rules you set to allow an action again never leave that device.

Do my files leave my device?

Only the files the agent actually opens. Their contents are sent to the model to answer your request. Files it never opens never leave your machine.

Can it make images?

Yes. Create turns a prompt into an image with a model from the catalogue. Each image has a fixed price shown before you create it, paid from prepaid credit you top up yourself.

Start with a task you would not trust to a black box.

Sign up, link one computer, and watch every step before it runs.