Trust is built,
not claimed.

Autonomy ladder

Authority is earned.

1. Watch

   Read-only. It reads, analyses, advises. It changes nothing.

2. Draft

   It prepares work for review. Nothing is sent or published.

3. Ask

   It acts after your approval, one request at a time.

4. Act alone

   Only for task types where its record shows it agrees with you and gets results. You can take it back at any time.

Hard limits

What it cannot do.

• Money. New spend and purchases need your approval.
• Public and irreversible. Publishing, sending and deleting need your approval.
• Risk levels are rules, enforced by software, not left to the AI's judgment.
• A kill switch. One file stops every outward action.
• Untrusted content is data. Web pages and documents are never treated as instructions.
• Credentials stay out of shared files, and the AI does not read them.

Your data

The memory is yours.

Company knowledge lives in plain, readable files you own and can export. The AI model underneath can be replaced without losing your history. Every action is written to a log you can read.

Known limitations

Where we fall short.

• ARSLAN is at the design-partner stage. We start in shadow mode, read-only.
• It can be wrong. That is why every recommendation carries the case against it and results are measured.
• We have no independent security audit or certification yet.
• We have no published results yet. See Evidence.
• It does not replace your judgment, your accountant or your lawyer.

Ask us how it is constrained.

We work with a small group of owners. You lead. ARSLAN challenges, executes and remembers.