Otter.ai alternative: doing meeting notes locally on a Mac

I build Outlier, so read this as an interested party's comparison — though I point at primary sources for every number. Otter is a purpose-built meeting product, and for a lot of people it's the correct answer. This page is about a narrower question: which parts of AI meeting notes actually need a cloud service, and which parts don't.

What Otter.ai is

Otter is a US-based, closed-source subscription service. It transcribes meeting audio with speaker labels on Otter's servers, then generates summaries, action items and searchable AI chat across your whole transcript library. Its bot joins Zoom, Google Meet and Microsoft Teams from your calendar — including meetings you skip — and the Mac/Windows desktop app, announced in October 2025, can record bot-free instead of sending a bot into the call.

As of July 2026: Basic $0, Pro $16.99/user/month monthly or $8.33 annually, Business $30 or $19.99 annually, Enterprise custom. The caps matter more than the prices. Basic gives 300 recording minutes per user per month, 30 minutes per conversation and three lifetime file imports; Pro raises that to 1,200 minutes a month, 90 minutes per meeting and 10 imports. Unlimited minutes and four-hour meetings start at Business. API and webhooks are Enterprise-only, SSO/SCIM is listed as needing a 100-user minimum, HIPAA is an Enterprise add-on, and an Otter MCP server is included from the free plan up. Live transcription covers six languages, on web, macOS, Windows, iOS, Android and a Chrome extension.

Where the data goes

All computation and storage is cloud-side: Otter's security page describes data in AWS S3 with AES-256 server-side encryption, and no on-device or offline mode is documented anywhere on Otter's own pages. What leaves the machine includes audio, uploaded media, transcripts and automatic OtterPilot screenshots, plus account and device identifiers, IP address and inferred location.

Two Otter pages sound contradictory on training until you notice the scoping. The privacy policy, effective 16 June 2026, says Otter trains "our proprietary AI technology on de-identified audio recordings and on transcriptions," and that data labeling providers receive shared data to build training sets. The security page's prominent no-training promise is scoped to third-party AI vendors — the subprocessor list names Anthropic and OpenAI — and imported files are excluded as well. Otter says it de-identifies first, and that human review of audio requires explicit customer consent. On the primary pages I could read, I found no described self-serve opt-out from Otter training on your de-identified recordings; the opt-outs the policy describes cover sale/sharing and targeted advertising. Help-center pages reportedly describe a training toggle, but they return 403 to automated fetches — unconfirmed.

Every subprocessor Otter lists is US-based; no EU data-residency option appears on Otter's own pages, and third-party analyses say there isn't one even on Enterprise. Consent risk sits with whoever deployed the recorder: third-party reporting describes a consolidated federal class action in California alleging wiretap and CIPA violations for recording without all-party consent, unresolved as of mid-2026. I couldn't fetch the docket, so treat that as reported, not verified — and note that in two-party-consent states, getting everyone's agreement is your job.

What Outlier is, and what it can't do

Outlier is a native macOS app for Apple Silicon (M1 or later, macOS 12+). Models run fully on-device across seven tiers, Nano 4B up to Plus 397B-a17b. No account, no caps, works with Wi-Fi off, nothing uploaded. Nano and Lite are free; Pro is $20/month or $149/year, with lifetime licenses from $99.

Now the part a comparison page usually buries. Outlier doesn't record meetings, doesn't join calls, and doesn't do speech-to-text. No bot, no calendar integration, no mobile app, no Windows build, no cloud sync, no shared workspace. If you want notes to appear on their own after a meeting, Otter does that and Outlier doesn't.

The local version is two steps: get a transcript some other way — on-device speech-to-text, or one your conferencing platform already produced — then hand that text to Outlier for the summary, the decisions and the follow-ups. That's a manual seam Otter doesn't have. It's also where the actual thinking happens.

Side by side

Where Otter genuinely wins

These aren't reluctant concessions. They're the parts that genuinely need a server.

• Zero local compute. A four-hour multi-speaker meeting transcribes without touching your CPU, GPU, RAM or battery, and works the same on a base MacBook Air or a phone. A local model's quality is bounded by the machine it runs on; Otter's isn't.
• Capture is solved end to end. The bot joins from your calendar and records meetings you aren't even in. Any local approach depends on you being there and hitting record.
• Live transcription with speaker labels while the meeting is still running — what makes late-joiner catch-up and in-meeting search possible at all.
• Built for more than one person, on every device: web, macOS, Windows, iOS, Android and a browser extension over one shared library, with permissions, comments, admin controls, retention policies and SSO/SCIM — a whole category a single-user Mac app doesn't serve.
• Corpus-wide search and downstream plumbing. AI Chat queries every meeting you've had, not one file at a time, and results flow into Salesforce, HubSpot, Slack, Jira, Notion and Google Workspace, an Enterprise REST API, and a hosted MCP server available even on the free plan.
• Compliance paperwork that unblocks purchases: SOC 2 Type 2, a published subprocessor list, admin-set retention, HIPAA as an Enterprise add-on.
• Purpose-built speech models. Otter's own ASR and speaker-ID models are trained on enormous volumes of meeting audio across six languages and a wide accent range. A general-purpose chat model is no substitute, and I won't pretend otherwise.

What you gain locally

The trade is narrow but real. The recording is the sensitive artifact — a board discussion, a privileged call with counsel, a diligence session under NDA. Keep it on your own machine and it never becomes someone else's stored object, never enters a training corpus, and never waits out a retention policy. There's no toggle to hunt for because there's no upload. You also stop metering, and the summarizing holds up: on a 54-prompt head-to-head, Core 27B matched Claude Opus on 98.9% of rubric checks.

Who should pick which

Pick Otter if you need meetings captured automatically, you need live transcription during the call, you're on Windows or a phone, you're buying for a team that shares a transcript library, or you need an API, SSO or a signed DPA. If any of those is a requirement, this isn't close.

Pick a local setup if the recordings themselves are the confidential thing, you're one person rather than a team, you're on an Apple Silicon Mac, or you'd rather not pay per seat forever. You're accepting a manual step in exchange for the audio never leaving your machine.

Plenty of people should use both: Otter for routine standups, local for the conversations that shouldn't be uploaded at all.