How it works
The short version: the model runs on a server in your office, and every agent is governed like a new employee.
The server runs open-weight models locally on Apple Silicon. Requests from your team go to that server, not to a cloud AI provider.
Your team
Macs, iPhones, Apple Watch and the tools you already use
Your network
Your office network, or your own Tailscale network away from the office
AIC server
Izuran Pro: the models and the agents
Conversations and logs stay on it
Your files
The folders you choose
The only outside connections
Models whose weights can be downloaded and run on your own hardware. We choose and test models for your use case. Every model we offer is tested on agent tasks in our lab before it reaches a customer.
Which ones fit depends on the size of your server. Models by size →
You wouldn't give a new hire every key on day one. Agents get the same treatment.
Each conversation works in its project's folder, and changes outside it need approval. Some actions always ask first, like deleting a calendar event or sending mail. Private projects keep their conversations on an encrypted drive.
Approvals are decided by fixed rules, not by the model. A payment, a new website, a change outside the project or anything near stored passwords needs a person's approval, with Touch ID, Face ID or a password.
Every conversation is kept as a log on the server, and a network log shows what connected where. Your team can open it in Settings.
Agents prepare, draft and check. An approval shows up on your Mac, iPhone and Apple Watch at once, and the first answer settles it.
“Trust the controls, not the explanation.”
Malek Ould-Oulhadj, founder
Running AI locally helps keep sensitive data in your control. It doesn't make you HIPAA, SOC 2 or any other standard compliant on its own, and we don't hold those certifications.