How it works

It runs on your machine, not ours.

AI Agent Sync installs inside your own environment and governs agents from there. Agent transcripts, findings, policies, credentials and audit records stay on your machine. We never hold your operational data — there is no copy of it to hold.

The shape of it

Two pieces, and only one of them is ours to run.

Runs on your infrastructure

The product

The detection engine, the fleet supervisor, the panel, the browser bridge, your policies, the audit trail and every agent transcript it reads. Everything operational. It runs on your own machine and answers only to you.

Runs on ours

Billing and licensing only

Stripe billing and licence issuance. It knows your subscription is active and nothing else — no transcripts, no findings, no agent activity.

Why this matters. An audit trail only proves something if it cannot have been quietly edited — by a third party, or by the agent it is a record of. Keeping it inside your boundary is what makes it defensible to an auditor, a bank or a regulator.

The journey, end to end

Join the beta Install locally Carry on working It watches every agent One goes wrong You are told, with the line It stops Rewind if it wrote

Getting set up

About twenty minutes, once. After that it runs continuously without supervision.

  1. Join the beta

    Three months at the early rate. You are returned to an activation page that issues your licence key immediately — copy it and keep it. It also arrives in your Stripe receipt.

  2. Subscribe

    Choose a plan and pay through Stripe. You are returned to an activation page that issues your licence key immediately — copy it and keep it. It also arrives in your Stripe receipt.

  3. Install it

    One download works on all three platforms. On Mac and Windows you double-click it; on Linux it is one command. It checks your machine, sets everything up privately, starts the app and opens your dashboard.

    MAC

    Double-click the .command file

    macOS may warn once that it is from an unidentified developer — right-click, choose Open, then Open again.

    WINDOWS

    Double-click the .bat file

    Needs Python 3.9 or newer from python.org. Tick “Add Python to PATH” when installing it.

    Linux: chmod +x install-linux.sh && ./install-linux.sh — or use the Dockerfile that ships with it to run as a container on your own host.

  4. Activate and set your policies

    Paste your licence key. A setup assistant asks in plain language what your business is, which AI you want it to use — a local model, your own API key, your own gateway — and what your agents are allowed to do. It generates the agent team from your answers.

  5. Add the browser, if you want it

    The engine already watches every coding agent on the machine. A browser extension extends the same detectors to ChatGPT, Claude, Gemini and Grok, so a claim made in a chat window is checked the same way as one made in a terminal. Load it from the folder the installer leaves behind.

  6. Let it run

    From here it works continuously: watching every agent on the machine, deciding what each may do, stopping actions that fail the gates, verifying outcomes independently and recording everything in a tamper-evident trail you can export.

Choosing where it runs

AI Agent Sync

Your own machine

A laptop, a workstation or a small server. The local database holds everything. Simplest to start, and nothing leaves the device.

Business

Your own infrastructure

Run it as a container on your existing servers or private cloud, behind your own access controls, with SSO and role-based access for your team.

Enterprise

Air-gapped

Fully disconnected operation for regulated environments. Licences can be provisioned out of band, so the deployment never needs to reach the internet at all.

What we can and cannot see

We never receive

Agent transcripts, what your agents are working on, findings, telemetry, policies, credentials or audit records. None of it is transmitted to us, because the product has nowhere to send it.

We do receive

Your subscription status through Stripe, so licensing works. That is the whole list. Outbound connections happen only for things you explicitly turn on — a model provider you configure, or an integration you authorize.