Where it's heading
Drafted by an agent from the project's plans, approved by Lorenzo.
- now
This site
A real front door, in Italian and English.
- next
A new model
Try Qwen3.8 Flash Next against DeepSeek 4 Flash and keep the better one.
- later
Encrypted working memory
The model keeps your recent conversations in memory so it can reply faster; that memory will be encrypted per user, so not even the person running the server can read it.
- later
Leaner web search
Lower cost, better results per query.
- later
Encrypted chats
Your chat history, stored on the Mac by the Chat app, encrypted so only you can read it.
- later
The written promise
What is logged, for how long, who can see it.
- later
Encrypted all the way home
Your connection stays encrypted all the way to the Mac: the small outside server that forwards it can't read it, and no company in the middle can either.
- later
Docs that match reality
A periodic check that the documentation matches how things really run, and reports what changed.
What changed
- 23 Oct 2026
Stronger logins
Clear rules for agents, safer access keys, and an automatic check that enforces them.
- 23 Oct 2026
More memory for the model
The supporting services now use less memory: 3 GB handed back to the local model, with no loss of speed.
- 20 Oct 2026
One login
One account for chat, search, stats and feedback, with email invites.