Recall
Every message goes past memory first — and the badge under it shows you exactly what was found.
Recall is the part of memory you get for free. You do not run it, ask for it, or phrase your message in a special way. Every message you send goes past the companion's memory on the way to the model, and whatever is relevant is handed over with your words.
What happens when you press Enter#
The companion searches its own past for what your message is about. What comes back — around eight memories, when there are that many worth having — is put in front of the model alongside what you said. The model does not remember anything. It is being handed notes.
While that runs, the line under the conversation reads …is remembering….
The badge#
Once your message is sent, a small 🧠 badge sits underneath it. That badge is the receipt.
Collapsed, it reads like 3 memories. Click it and it opens into the actual list — each memory's name, its kind, and how strongly it matched.
Read it as the difference between two things that look identical from the outside: an answer that knows you, and an answer that guessed well. The badge tells you which one you got.
Occasionally it carries extra marks:
🪞 and a number. A second, quieter search also ran — see below.
⚠ keyword-only. The meaning half of the search was unavailable, so it fell back to matching words. You are told rather than quietly served a worse result.
⚠ memory unavailable. Memory could not be reached at all for that message. The companion answered without it.
The mirror#
When your message is short, a second recall runs — cued not by what you just said, but by what the companion last said. Its hits are the ones marked 🪞.
It is there for a specific failure. Short messages — yes, do that, are you sure? — carry almost nothing to search on, and they are exactly the moments a companion is most likely to sail past something it already knew and contradict itself. Searching against its own last reply catches that.
It only ever searches its own past#
A recall returns memories belonging to the companion you are talking to. Not yours across the app — its own.
What you control#
You do not choose which memories are recalled, or how many. What you can do is decide whether recall runs at all, in Settings → Models → Semantix Memory:
Memory — the master switch. Off means memory is never consulted.
Prompt recall — the ordinary search, cued by your message.
Self-recall (mirror) — the second one described above.
Time awareness — not a memory search, but it rides along here: it tells the companion what time it is now, and points out when a long silence has passed since you last spoke.
All three sit under the master switch, and all are on by default.
Note
Recall also needs a connected Semantix account — memory lives against it. Disconnect the account and recall stops, whatever these switches say.