Claude Minutes Conversation Memory connectorWhat Claude can do with the meetings recorded on your machine.

Tells Claude whether a recording is currently in progress, a quick check before starting another one or asking about the last meeting. It only reports the recording state and starts or stops nothing.

Lists the background processing jobs for recent recordings, so Claude can say which transcripts are done and which are still being worked on.

Returns your recent normal meetings and voice memos. Restricted items are left out unless an explicit, logged override is both enabled and requested. Restricted items need all of the override conditions before they appear.

Looks through the transcripts of normal meetings and your voice memos for a topic or phrase, with restricted items excluded under the same override rule.

Retrieves the transcript of one normal meeting. For a restricted meeting, Minutes returns a content-free stub instead, unless the logged override is enabled and requested.

Summarizes the desktop context bound to one exact normal meeting, so Claude can relay what the desktop context recorded around that conversation. Its scope is one exact meeting source, not your whole desktop history.

Searches the desktop-context events tied to one exact normal meeting, rather than across everything you did that day. The search stays bound to that single meeting source.

Shows the local desktop-context rewind bound to one exact normal meeting, a way back to what was happening at a given point of that conversation.

Brings back bounded, verified screenshots tied to one exact normal meeting, so Claude can refer to what was visible rather than guess from the transcript.

Flags conflicting decisions and stale commitments across your meetings, the kind of drift that is hard to notice from one conversation to the next. It flags the problems and leaves every decision and commitment as recorded.

Builds a profile of someone from policy-authorized meetings, within the corpus bounds Minutes supports, so Claude can brief you on past exchanges with them.

Researches one topic across policy-authorized meetings, within supported bounds, and gathers what was said about it over time. Restricted meetings stay outside those authorized sources by default.

Lists open and stale commitments from your meetings, optionally filtered by person, so promises made out loud do not quietly disappear. Without the person filter, every open or stale commitment comes back.

Ranks relationships from a bounded, process-private, policy-safe graph projection, giving Claude a sense of who you talk to most. It ranks relationships and leaves your meetings as they are.

Lists the voice profiles enrolled for speaker identification, so Claude knows which speakers Minutes can recognize in transcripts. Confirming a speaker is not something Claude can do: Minutes keeps that in its app or command line.

Queries decisions, commitments and questions extracted from meetings. Minutes re-verifies each source meeting against the live sensitivity policy before release and reports withheld records as a partial view.

Reads utterances from the active live transcript, with an optional cursor or time window, so Claude can catch up on the latest part of the conversation.

Reads the current copilot session and the health of its provider, so Claude can say whether the copilot is running properly. The information comes from the copilot's command-line status surface.

Reads the nudges the copilot produced, step by step through a cursor or a time window, so they can be reviewed afterwards. It reads nudges already observed and does not start the copilot.

Shows the current state of the knowledge base: its configuration, adapter, people count and log entries. That knowledge base is the one ingest_meeting feeds with person profiles, a log and an index.