01 What you get

• Operator installed on your Linux machine or server. In your office or in your own cloud account. The agents, the workflows, the logs and the vault run there.
• Your systems connected. We set up connectors for the tools you use, typically email, WhatsApp, your CRM or ERP, spreadsheets and internal APIs.
• Your workflows configured. What starts each one, what the agent may read or change, who approves, and where the result goes.
• An admin panel where you see every run, approve or reject waiting actions, manage users and connectors, and read the audit log.

02 Connectors: what the AI is allowed to do

A connector turns one of your systems into a short list of defined actions, such as read order, update contact or send message. Each action is marked by risk: read, change or destructive. Destructive actions always need a person to approve them.

It uses your company's own accounts on that system. The agent never calls your systems directly: it asks for an action, Operator checks that this workflow may use it, holds it for approval if your rules say so, then the connector makes the call with the right credential. Every call is logged.

03 The vault: why the AI never needs your passwords

To be useful, automation has to log in to things. The usual shortcut is pasting API keys and passwords into an AI tool or a config file. Operator doesn't work that way.

Credentials are stored encrypted (AES-256-GCM) in a vault on your machine. When an approved action runs, the connector adds the credential to that one outgoing call. The agent only ever sees the action and its result, and known secret values are scrubbed from results and logs. Even the AI provider key is added by Operator, not handed to the agent.

So you can let agents work across your systems without putting secrets in prompts, chats or scripts, and you can revoke or rotate a credential in one place.

04 An example workflow

1. In: a customer asks on WhatsApp where their order is.
2. Look up: the agent uses the ERP connector's read-only find order action.
3. Prepare: it drafts the reply and a note for the CRM.
4. Approve: someone on your team sees exactly what will be sent and approves or rejects it.
5. Out: the reply is sent, the CRM is updated, and every step is in the log.

An illustration. Which systems connect out of the box and which connectors we build during setup depends on your stack; we confirm that before we start. Each workflow can be draft-only, approval-required or autonomous, and you decide when it moves up.

05 Where your data goes

Operator, its database, the vault and the run history live on your machine, under your control. We don't copy your data. If the installation is linked to us for updates, it sends us operational metadata only (run counts, errors, versions, cost), and you can see a log of what was sent. Support access is a time-limited session that you open.

Your data still goes where a task needs it: the AI provider you choose receives the content of each request, and your business services receive the calls made to them, under your own accounts. You choose the providers; Claude, ChatGPT and other agents run under your accounts.

06 How you pay

• Setup fee for installation and configuring your systems and workflows.
• Monthly fee per deployment while we keep it updated and supported.
• Buyout option: after 12 months you can buy out the installed version, or agree an upfront buyout as part of the project.

A buyout gives you a perpetual, non-exclusive license to the version installed. Your workflows, connectors and data are yours. The Operator platform itself stays ours, and updates and support after a buyout are only included if we agree them separately. Prices are quoted per project.