Meeting tools can produce summaries. They are less good at knowing what mattered to you at the moment someone said it.
Trace, a local macOS transcription app, adds a very small gesture to that workflow. During a call, a keyboard shortcut marks an instant and lets you attach a note. The marker then appears directly inside the transcript at the right timestamp.
It is not the product's flashiest feature. It may be the most useful one.
The human leaves a signal while the context still exists
An automatic summary works after the fact. It has to infer which passages mattered from the transcript alone.
Trace lets the person who is actually in the meeting press a key when a decision, disagreement or detail is worth keeping. There is no need to open a separate notebook and reconstruct the context twenty minutes later.
The final transcript is Markdown. Flagged moments are useful to a human reader, but they also survive when the file is later handed to a model.
It is a small form of human supervision that does not require reviewing the entire output.
The rest stays local
The current Trace captures microphone and system audio, can separate speakers, and produces both transcripts and summaries on-device. Its website says recordings are never uploaded and no account is required.
It requires macOS 14.4 or later and Apple silicon. It can also import audio or video, search previous sessions and send Markdown to tools such as Obsidian or Bear.
The product has changed substantially since its first Show HN in June. Back then, its maker explicitly said Trace did not perform summarization itself. That claim is now outdated: the current site documents local summaries and transcript-grounded answers.
That is a useful reminder about ordinary software fact-checking. Two months is enough for an article to describe the old product.
The shortcut ages better. Instead of asking AI to guess everything that matters, Trace lets the human drop a few breadcrumbs while the conversation is still alive.