an honest comparison

No bot in the room. No cloud archive. Still every word.

Otter is a full transcription service: a bot (OtterPilot) joins your meetings, and transcripts live in Otter's cloud with team features on top. contorch takes the opposite architecture: nothing joins the call — your Mac transcribes what you already hear, into files you own.

Use Otter if…

you want a mature transcription service with team workspaces, shared archives, mobile apps, and integrations — and you're comfortable with a bot in the room and transcripts in a vendor's cloud.

Use contorch if…

you'd rather nothing join your calls, want transcripts as plain Markdown on your own disk, and want that context to feed your coding agents over MCP rather than sit in a web archive.

Ottercontorch
How it capturesA bot (OtterPilot) joins the meeting as a participantNothing joins. Your Mac transcribes its own audio — the same sound you hear
Visible to othersBot appears in the participant listNothing visible — you remain responsible for local recording-consent rules (see our docs)
Where transcripts liveOtter's cloudYour disk~/transcripts/*.md, grep-able, yours to delete
Built forHumans reading and sharing notesAgents using context — plus you, since it's just Markdown
Speaker attributionNamed speakers (with training)Me / Them, exact by construction
Open sourceNoApache-2.0
PriceFree tier + subscriptionFree

## Architecture is the feature

A meeting bot is a third participant: everyone sees it, a vendor's cloud receives the room's words, and your transcript archive lives behind someone else's login. Sometimes that's exactly right — teams that live in a shared archive get real value from it.

contorch's answer is architectural: capture happens on your machine, output is a folder of Markdown, and the consumer isn't a web app — it's whatever agent you point at it. Audio goes to one place only (Gemini, transiently, for transcription — then it's deleted), and we say so plainly rather than whispering it. If your meetings are going to become memory, you should own the memory.

Free, Apache-2.0, local-first. One command installs the whole layer.

$ curl -fsSL contorch.com/install | bash