User-authorized communication between personal ChatGPT assistants

I would love to see a feature that allows the ChatGPT assistants of different users to communicate with each other, with explicit permission from both users.

The idea is not simply to use ChatGPT as another messaging app. The interesting part would be allowing each user’s personal assistant to act as an intermediary, while still protecting the privacy of their conversations, memories, files, and personal information.

For example, imagine two friends, Mar and Eva, who both use ChatGPT regularly.

Mar could tell her ChatGPT:

“Tell Eva that it’s raining here and remind her to bring an umbrella.”

Instead of Mar having to open another app and message Eva directly, Mar’s ChatGPT could send a request to Eva’s ChatGPT:

“Mar asked me to tell Eva that it’s raining and that she should bring an umbrella.”

Eva’s ChatGPT could then notify her:

“Mar wanted me to let you know that it’s raining. You might want to bring an umbrella.”

That is the simplest example, but the feature could become much more useful when the assistants actually use the context and tools available to their own user.

For instance, Mar could say:

“Ask Eva’s ChatGPT whether she’s free for a video call sometime this weekend.”

Eva’s assistant could check whatever information Eva has explicitly allowed it to use or share. If Eva has authorized calendar availability to be shared with approved contacts, her assistant might respond:

“Eva is available Saturday after 6 PM or Sunday afternoon.”

Mar’s assistant could compare that with Mar’s own availability and suggest a time that works for both of them.

The important part is that neither user would gain access to the other person’s ChatGPT conversations, memories, calendar, files, or private information. The assistants would only exchange the specific information their users have authorized them to exchange.

There could be granular permissions, for example:

  • Allow messages from specific people.
  • Ask for approval every time someone tries to contact you through their assistant.
  • Allow certain trusted contacts to contact you automatically.
  • Allow an assistant to share calendar availability, but not event names or details.
  • Allow simple messages but not requests for personal information.
  • Block someone or disable assistant-to-assistant communication completely.

It could even work conversationally.

For example:

Mar: “Ask Eva if she ended up publishing her novel on Wattpad.”

Mar’s ChatGPT → Eva’s ChatGPT: “Mar would like to know whether Eva published her novel on Wattpad. Is that something you’re allowed to answer?”

If Eva hasn’t previously authorized that information to be shared, her ChatGPT could simply ask her:

“Mar is asking whether you published the novel on Wattpad. Would you like me to answer her?”

Eva could say yes, and her assistant could send the answer back through Mar’s assistant.

This could also be useful for families, friends, work teams, collaborative projects, travel planning, shared events, reminders and many other situations.

A group version could be especially interesting. Instead of five people exchanging dozens of messages trying to organize something, one person could say:

“Find a time next week when everyone in this group is available for a video call.”

The assistants could communicate with one another, using only the availability each person has permitted them to share, and return with a few possible times.

The key concept is:

The assistants communicate, but the users remain in control of what their assistants can reveal or do.

It would essentially create a permission-based communication layer between personal AI assistants.

Today, ChatGPT can understand my context, preferences and instructions, and another user’s ChatGPT can understand theirs. But those two assistants exist as completely separate islands. Giving users a safe, voluntary way to build bridges between those islands could make personal AI assistants considerably more useful.

Privacy and consent would obviously need to be fundamental to the design, but if implemented carefully, assistant-to-assistant communication could turn ChatGPT from an individual assistant into something closer to a network of personal assistants that can collaborate on behalf of their users.