AI Passport vs a Shared Memory Layer: What Actually Works for Scheduled Agents

Quick answer: AI Passport (Egoist Machines, launched October 2026 with a $3M seed round that included Y Combinator and Goodwater Capital) is a permission-based portability layer for your personal data: tell ChatGPT something and Claude can know it, but only after the transfer passes through an inbox you review, with per-request permission and sensitive details like dates of birth never shared without explicit approval. That design assumes a human is at the keyboard. Your scheduled agents have no human at the keyboard. A shared memory layer is built for the opposite case: every AI client connects to one account over MCP, everything it saves (decisions, rules, full conversation history) is readable from every other connected tool, and the newest correction becomes the truth everywhere, with no approval queue. Pick by the setting: personal data with a gate, or agent working context with no gate at all.

What AI Passport actually does

The model is explicit in the launch coverage: AI Passport connects to about 800 applications (ChatGPT, Claude, Gemini, Perplexity, plus productivity tools like Notion and Gmail and wearable devices, with more added daily). When you add information in a connected app, it syncs to a user-owned inbox. You periodically review the inbox and decide per item whether it can be shared with other apps, stays only in the original app, or gets deleted entirely.

Connected apps can also request information about you. You set defaults for accepting requests, but sensitive fields like a date of birth are never handed out without explicit permission. Nothing moves without your say-so, and the company is explicit that this is the point: CEO Erin McGurk framed it as a permission-based system for sharing personal data, where users stay in control of what their AI assistants know about them.

The honest tradeoff of the passport model: you are the bottleneck. Every transfer waits on your review. For personal data where control is the whole point, that is the feature. For working context that changes by the hour, the same control becomes a queue you have to keep clearing.

The scenario it was not designed for

Now put a scheduled agent into that picture. A lead-triage workflow fires at 6 AM in your automation platform. It reads the overnight leads, scores them, drafts follow-ups, and hands the results to a second agent that books the meetings. Nothing in that chain has a human watching. There is no one to open an inbox, review the transfers, and approve them before the 6:05 AM booking agent needs the context the 6:00 AM triage agent just produced.

This is not a flaw in the passport design. It is a category difference. Permission-gated portability was built for interactive assistants with a person sitting in front of them: you chat with ChatGPT, your travel plans land in your inbox, you approve them, and Claude can use them the next time you chat. Scheduled agents invert every one of those assumptions. They run unattended, they produce context continuously, and they need it back within seconds, not after someone clears a queue.

So the question for an automation operator is not whether AI Passport works. It is whether anything with a human approval gate can work inside a loop that runs when you are asleep. The answer is no. The gate has to go, and the context has to live somewhere the agents can reach directly.

What a shared memory layer actually does

A shared memory layer skips the inbox and the approvals. Each AI client connects to one account over MCP, and everything it saves (exchanges, decisions, tasks, rules) is readable from every other connected tool. When two tools save conflicting information, the most recent save becomes the truth, so you correct something once and every tool sees it.

One example of this kind is Vilix AI: cloud-hosted, so there is zero infrastructure to manage; the same memory follows you across phone apps, laptop and every AI tool; it stores full conversation history rather than just extracted facts; and your data is portable, with export and delete anytime. Some users have connected 60+ tools and run scheduled agents that read and write daily, with the newest correction becoming the truth everywhere. There is a free plan forever, plus a 7-day Pro trial that needs no credit card.

The practical consequence for the 6 AM workflow: the triage agent saves its scoring decisions to the shared layer as it works, and the booking agent reads them at 6:05 without anyone approving a transfer. The context moves at the speed of the automation, not the speed of your inbox review.

The side-by-side

How to choose without overthinking it

Ask one question: is there a human in the loop when the context needs to move?

If the answer is yes, because you are chatting with assistants and the whole point is that nothing crosses tools without your say-so, you want the passport model. Sign up at ego.ist, connect your apps, review the inbox, and treat the occasional review queue as the price of control.

If the answer is no, because your agents run on schedules and the whole point is that they never start blank again, you want the shared layer. Connect the tools you actually use, correct things once, and let the newest truth follow your work from the laptop to the scheduled run to the phone.

The confusion is that both are called "sharing context." A passport shares your data with apps. A memory layer shares your work with your tools. They solve different problems, and picking the wrong one gives you either an approval queue where you needed zero friction, or zero friction where you needed an approval gate.

Give your scheduled agents one memory they all read. Get Started for Free Free forever, no credit card.