In development
Designed to ask before it acts.
Safety is part of how Nadeemi is being built, not added on top. This is the design we are building to.
What is on the always-ask list?
Sending to anyone new, sharing a file, deleting anything and any step that costs money. These will need your approval on your phone with a passkey, and an approval will expire after 24 hours. The list cannot be switched off by accident or by a message.
What are permission modes?
You choose how much the agent may do alone: ask every time, ask only for the always-ask list, or allow a fixed set of actions for one task. You can change the mode at any time.
What will happen when I press stop?
Work stops within seconds and nothing new starts. We will also have our own stop, and we plan to test both regularly.
Will I see every step it took?
Yes. Every step will be written to a tamper-evident record that you can export.
What if someone tries to trick it by email?
Incoming mail will first be read by a separate reader that has no tools and cannot act; it can only report what the mail says. The agent will contact only approved addresses, and new recipients will always need your approval. Each release will be tested against Arabic and English attack attempts before it ships.
Want to hear when invitations open?
Today you can try Nadeemi Beta, a pay-as-you-go AI chat. For the full agent, request a private invitation by email and we will write when invitations open.