How we use AI
On this site
No AI model reads anything on this site. “Send it to the office” matches your words in your browser with plain rules, and every message you send us reaches a person: Ali reads each one.
In the offices we build
The agents run Claude Code, from Anthropic. We choose the model for each job, and the office records which model did every piece of work.
These are the services that touch your data, and why:
| Service | What it sees | Why |
|---|---|---|
| Anthropic (Claude) | The requests and the parts of your code or documents a task needs | To do the work |
| Your chat channel (Telegram or Zoho Cliq) | Your requests and the office’s replies | So you can run the office from your phone |
| Your own server, in your name | Everything the office does, and its record | Where the office runs |
| Your own accounts (your code, your store) | What you give the office access to | Where the changes land |
The agents work in locked boxes that can reach only the AI provider. Every step is written down, and the record cannot be edited afterwards.
What AI never decides
An agent does not start a task, approve a plan that touches payments, sign-in or your data, or put anything live. Each of those waits for you: every task starts with your tap, and production takes three steps from you. A second agent that never saw the plan reviews the work before you see it.
When an agent gets it wrong
The office is built to catch it: thin requests go back to you, your project’s own checks run on every change, and a staging deploy that fails its checks rolls itself back.