Sayli runs a loop four times a second while you are on a call. You do not see the machinery. You see a window that knows what is happening and helps when you ask. Here is the loop, in plain English.
It listens
A Mac app listens to the call through your system audio and your microphone. Both streams stay on your machine until you are in a session. Nothing dials into the meeting. The host sees no extra participant, no recording banner, no sign that Sayli is there.
Because Sayli captures two channels, it can tell you apart from the other person. You are one voice. The call is the other. That separation is what makes the brief say who said what.
It transcribes
The words appear on Sayli's side as the call runs, usually under a second behind the speaker. Before any of that text reaches a language model, Sayli redacts the sensitive parts: card numbers, social security numbers, phone numbers, emails, and IP addresses. The model sees the conversation, not the secrets inside it.
PII is scrubbed before the text is stored, before it is shown, and before any model reads it. See Privacy and security for the full list and the reasoning.
It thinks
When you ask a question, Sayli gathers three things and reasons over them together:
The live thread of the conversation, so the answer fits what was just said.
The most relevant passages from the docs and notes you gave Sayli, so the answer cites your facts.
A live read from a connected service when it helps, like a Jira ticket or a Linear issue, pulled in over MCP.
The answer streams back token by token, so you can start reading before it finishes. Ask in your words. Sayli answers in the language you set, even if the call is happening in another.
It remembers
When the call ends, Sayli writes the brief: a headline, a short summary, the action items with owners, and a follow-up email. Every claim points back to the moment in the transcript where it was said.
Then the call joins your memory. Past conversations are chunked and indexed, so weeks later you can ask one question across everything you have discussed and get an answer grounded in what actually happened. Memory is the part that compounds. The product gets better the longer you use it.