What it recalls as you type

Every message you send takes a detour before it reaches the model. Semantix looks at what you wrote, finds what it already knows that bears on it, and sends both. You don't trigger this and you don't see it happen — the reply simply arrives already knowing things.

Three separate reflexes do this work, in Settings → General → Memory. All three are on by default, and they're independent: turn off the one you don't want and keep the rest.

They only run when Semantix Memory is on in Privacy and Local only is off. Without that, this whole section is inert.

Prompt recall#

The main one. It reads the message you just typed and looks for anything related in this project's memory.

This is what makes the assistant open a conversation already knowing that you use pnpm, that the staging box is the one with the odd DNS, or that you decided against that library in March. You ask a question; the answer arrives with the context attached.

Time awareness#

Tells the assistant what the actual date and time are, and points out when a long gap has passed since your last message.

This sounds small and isn't. A model with no clock has no idea whether your last exchange was two minutes or two weeks ago — so it will cheerfully say "just now" about something from last month, or carry on mid-thought as though you never left. With this on, a long silence is visible to it as a long silence.

Self-recall (mirror)#

Looks at the assistant's own last reply rather than your message.

It earns its place on short prompts. When you write "yes, do that" or "why?", there is almost nothing in your message to search on — but there is plenty in what the assistant just said. This reflex searches that instead, so a two-word reply still gets context.

Which memory it reads from#

Each project has its own memory, and by default a conversation reads the one belonging to the project you have open. That's almost always what you want and needs no thought.

When you do want something else, the model dropdown in the composer toolbar has a Memory row just under the search box. It lists the open project's own memory first, then the others you've built.

The choice is separate from the model — the same model can be pointed at a different memory — and it applies both ways: the row you pick is what gets recalled as you type, and it's also where /sleep files what it distils.

The row is hidden entirely when memory is off.

Asking instead of waiting#

The reflexes are the automatic path. You can also just ask, in the conversation, and the assistant will do it deliberately:

• "what do you know about how we handle auth here?" — it looks it up
• "remember that the staging DB is read-only" — it writes it down

These need the same two switches on as the reflexes do. When the master switch in General is off, the assistant loses these abilities too, not just the automatic ones — it is never offered them, and if it tries one anyway it gets back memory is disabled by your settings rather than silently reaching for your memory.

→ Filing a conversation with /sleep · What's stored, and where