Support

Something not working? Start here.

Sumsight is made by a very small team. Most of what goes wrong is on this page, and the first item catches almost everybody. If the answer isn't here, write to us and a person will read it.

hello@sumsight.app

Before anything else

What Sumsight needs to run.

  • macOS 26 or later. Sumsight is built on the on-device speech and language models that arrived with macOS 26. It will not install on anything older.
  • A Mac with Apple Silicon. M1 or later. Sumsight does not run on Intel Macs, and we would rather tell you that here than let you find out after downloading it.
  • Apple Intelligence, switched on. This one is not optional and it is not a soft requirement — if Apple Intelligence is not available on your Mac, Sumsight shows you a screen explaining how to turn it on and goes no further. See the first question below, because this is where most people get stuck.
  • Permission to record, from whichever source you use. macOS asks once for system-audio capture and once for the microphone. You only need the one you actually record with.
  • Several gigabytes of free disk space. The app itself is small. What takes room is the models — Apple Intelligence is a 2–3 GB Apple download, and Sumsight fetches up to about 3 GB more, on demand, over the life of its use. The full list is further down this page.

Languages in version 1.0: English, Spanish, French, German, Italian, and Brazilian Portuguese. Each transcript segment is tagged with the language it was spoken in, so a conversation that switches languages stays readable.

Getting started

Four steps, once.

  • Pick your source. System audio captures whatever the Mac is playing. A microphone or audio interface captures the room. You can use one at a time, and the app always shows you which one it is on.
  • Grant permission the first time. macOS asks once per source. If you dismiss the prompt you can grant it later in System Settings › Privacy & Security — and after granting it, quit Sumsight and open it again, because macOS only applies the change on relaunch.
  • Press Listen. The transcript appears as it goes.
  • Press Stop. Sumsight writes the summary. On a long session this takes a little while — it is running on your Mac, not in a datacenter.

Common problems

In roughly the order people hit them.

Apple Intelligence says it's unavailable, and my Mac supports it.

This one catches almost everybody, and it is not a Sumsight bug. Apple requires your Siri language and your Mac's system language to match, and both have to be languages Apple supports. If they differ, Apple Intelligence reports itself unavailable and Apple's own Settings screen does not explain why. Open System Settings › Apple Intelligence & Siri, match the two, and give macOS time to finish downloading its models — that download is two to three gigabytes and wants Wi-Fi and power. Sumsight rechecks automatically every thirty seconds while it waits, so you can leave the screen open.

The app won't open past the Apple Intelligence screen.

That screen is the app telling you it cannot do its job yet, not a crash. Sumsight's summaries are produced by Apple Intelligence, so it will not proceed until macOS reports it as available. Follow the answer above; the moment availability flips, Sumsight continues on its own without a relaunch.

There's no audio, or the transcript stays empty.

Check the source first — the app shows which one it is on. For system audio, something has to actually be playing, and Sumsight needs permission to capture it in System Settings › Privacy & Security. For a microphone, check that macOS has the right input device selected and that it isn't muted. If you granted permission while the app was already running, quit and reopen it.

I don't see speaker labels.

Speaker labels only exist for microphone and audio-interface recordings. Anything the Mac plays arrives as a single mixed stream, and there is no reliable way to separate the voices in a mix — so Sumsight doesn't guess, and it removes the labels rather than showing you ones it cannot stand behind. Record through a microphone and labels appear on every turn, including in every export.

The summary looks different from the last one.

Apple Intelligence writes your summaries. If a particular session defeats it — a transcript too long for its context window, or a language it does not support — Sumsight falls back rather than failing: first to a local model it downloads once, then to a classical method that is faster but plainer. The app labels every summary with which one produced it, so you are never guessing.

It's downloading something.

It's a model. See the next section for the full list and the sizes. Each downloads once, none of them is needed before you start, and none of them carries anything of yours outbound.

My language isn't in the list.

Version 1.0 ships English, Spanish, French, German, Italian, and Brazilian Portuguese. More are on the roadmap. If you need a specific one, tell us — it genuinely affects what we do next.

Where are my transcripts stored?

In Sumsight's own container on your Mac, in an ordinary local database, indexed by Spotlight so you can search them. There is no cloud copy and no sync. Back up your Mac and your transcripts are backed up; they are protected by your Mac's own security and nothing else.

Can I get the audio file?

For microphone and audio-interface recordings, yes. For system audio, no — Sumsight never writes system audio to disk at all, only the transcript. That is deliberate and it is not configurable.

How do I get the transcript out?

Plain text, SRT subtitles, Markdown written straight into an Obsidian vault, or a self-contained HTML file. Sumsight also ships Shortcuts actions, so you can start listening, stop, summarize, or export without touching the app.

What it downloads

Three things, and nothing of yours.

  • Apple's speech assets for each language you use — roughly 50–200 MB each, requested and managed by macOS itself.
  • A speaker-labelling model, about 600 MB, the first time you use speaker labels.
  • A local summarization model, about 2.5 GB, downloaded from Hugging Face, and only if Sumsight ever has to fall back to it.

Each downloads once. None of them is needed before you start — they arrive when you first use the feature that needs them. They are requests for a model file, not uploads: your audio, transcripts and summaries are never part of any network request Sumsight makes. The full detail is in the privacy notice.

Reporting a bug

Send it from inside the app.

Settings › Help › Send Feedback opens a message in your own email app with your Sumsight version, macOS version and hardware model already filled in. You can read and edit every word before you send it, and no transcript content is ever attached.

Performance and crash diagnostics stay on your Mac for seven days and are never uploaded. If you want to include one, Settings › Help › Open Diagnostics Folder shows you exactly what is there and you can attach it yourself.

Contact

A person, not a ticket queue.

Sumsight is made by a very small team and a person reads every message. If you are writing about a privacy request, say so in the subject line and we will handle it inside the window set out in the privacy notice.