What you can expect next.
vpchat is good, but not finished. Here is what we are busy with, and how we keep new capabilities from watering down the promise: the AI gets the context, you keep the identity.
Agent workspace
Agents that take on multi-step tasks: research, work through documents, assemble results. What you get back is not one answer, but a finished task.
The catch: an agent that opens files and operates tools on its own keeps pulling in new data. So the workspace gets a door, and our engine stands at that door. Everything going in and out passes through it. The model does the work and still never sees the real personal data. Not black magic, but not an afternoon's work either.
Web search
Current information from the web, right in the middle of the conversation.
Searching with placeholders would be silly, because no search engine on earth finds anything for “[PERSON_1]”. So the question goes out, but not the person behind it. “Can Ms Weber from Dortmund still cancel?” turns into a search about cancellation periods, and whatever comes back runs through our engine on the way in. The search engine learns what was asked. Not who it was about.
What are you missing?
Requests, praise, criticism, bug reports, half-baked ideas: all welcome at vpchat@mobilistics.de.