otter.ai alternatives that don’t upload your audio

Why people look beyond Otter.ai

Otter.ai is a capable product: it joins your Zoom, Google Meet, or Teams call as a bot, records it, and stores the audio and transcript in Otter's cloud. That architecture is also the dealbreaker for a growing group of users. If you handle client calls under NDA, patient or legal conversations, journalism sources, or anything covered by a data processing agreement, the question isn't whether Otter's transcripts are good — it's whether you're allowed to send the recording to a third-party server at all. Otter's free plan covers 300 transcription minutes per month; Pro costs $16.99 per month billed monthly, or $99.99 per year, as of June 2026. Every plan uploads your audio.

This list covers tools where transcription happens on your own hardware and the audio file never leaves your device. One entry (Granola) is included as a caveat because it is frequently recommended as a 'private' Otter alternative when it is only bot-free, not cloud-free. One entry (now recording) is our own product, disclosed up front and held to the same standard as the rest.

1. now recording — on-device transcription in your browser (our product)

Disclosure: nowrecording.com is our product. now recording records and transcribes meetings entirely in your browser using Whisper running on WebGPU, or in the iOS app. There is no meeting bot, no calendar access, and nothing to install on the web — you open the page and hit record. Because audio is captured locally from your microphone and system audio, it works identically with Zoom, Meet, Teams, or an in-person conversation. The privacy claim is verifiable rather than contractual: open your browser's DevTools network tab while recording and you will see no audio leaving the machine.

The free tier has no account requirement and no recording or transcription limits: unlimited live transcription in 29 languages, on-device speaker identification, a high-accuracy polish pass, Markdown export, and 3 AI summaries per month (10 if you sign in). Pro costs $8 per month, or $5 per month billed annually, with a 7-day free trial as of June 2026. Pro adds unlimited AI summaries, cloud backup and sync of transcript text (text only — audio is never stored on our servers), cross-device access, full-text search across meetings, and public share links. An 'Ask your meetings' Q&A feature is in development.

The one disclosed exception to local processing: if you request an AI summary, the transcript text is sent to an LLM API. Everything else stays on-device. Honest limitations: there is no video recording, no CRM or calendar integrations, no Android app yet, and team/admin features are thin compared to Otter or Granola. The browser version needs a WebGPU-capable browser (current Chrome or Edge) for the best models. If you need enterprise admin controls or video capture, a cloud tool will serve you better today.

2. Meetily — open-source and self-hosted

Meetily (Zackriya Solutions) is the open-source heavyweight of this category, with 12.7k GitHub stars as of June 2026. It is a Rust-based desktop app (Tauri, with a Next.js frontend) that captures meeting audio locally, transcribes with Whisper or Parakeet models, performs speaker diarization, and generates summaries through Ollama running local LLMs — or, optionally, through Claude, Groq, or any OpenAI-compatible endpoint you configure. The Community Edition is MIT-licensed and free; the maintainers state core functionality has zero cloud requirement.

Pricing: free and open source for the Community Edition. A separate Meetily PRO tier exists for teams wanting enhanced accuracy, custom templates, and supported self-hosted deployment. Platforms: macOS and Windows have prebuilt installers; Linux requires building from source.

Best for: developers, self-hosters, and enterprises that want to audit the code and own the entire stack. Honest limitations: fully local summarization requires hardware capable of running an LLM through Ollama, setup is more involved than any commercial option here, and the interface is rougher than Otter's or Granola's. There is no managed mobile app.

3. MacWhisper — pay once, transcribe forever on your Mac

MacWhisper (by Good Snooze) is a native Mac app that runs Whisper models entirely on your machine. It transcribes audio and video files, records system audio for meetings, supports batch processing, and exports to common subtitle and text formats. Nothing is uploaded unless you explicitly opt into a cloud service for specific features.

Pricing as of June 2026: a functional free version, and a Pro license at €59 one-time (roughly $69) via Gumroad, with 25% student/journalist/nonprofit discounts on request. A Mac App Store variant called Whisper Transcription offers the same engine as a subscription: $6.99/month, $29.99/year, or $99.99 lifetime.

Best for: Mac users who refuse subscriptions and transcribe regularly — the one-time price undercuts four months of Otter Pro. Honest limitations: macOS only, no mobile companion, no cloud sync or cross-device search, and it is a transcription tool rather than a meeting workspace — organizing, searching, and sharing notes across many meetings is on you.

4. Superwhisper — dictation-first, meetings included

Superwhisper is primarily a system-wide voice-to-text tool: you talk, and accurate text appears in any app — Slack, Gmail, your IDE. It runs local Whisper-class models and works fully offline. It also includes a meeting assistant that records and transcribes meetings with automatic notes, which makes it a reasonable Otter replacement if dictation is your main use case anyway. As of 2026 it covers macOS, Windows, and iOS under a single license.

Pricing as of June 2026: a free tier with voice-to-text in any app, 100+ language support, and basic meeting transcription; Pro at $8.49/month or $84.99/year; and a $249.99 lifetime license. Students get 40% off, and paid plans carry a 30-day refund window.

Best for: people who want one tool for dictation everywhere and occasional meeting capture. Honest limitations: meetings are a secondary feature, not the product's center of gravity — there is no meeting library with cross-meeting search or sharing comparable to dedicated meeting tools, and the most capable AI formatting modes use cloud LLMs you configure.

5. Alter — a local AI assistant for Mac that also takes meeting notes

Alter positions itself as a native AI assistant for macOS, with meeting recording as one feature among many (workflow automations, document chat, dictation). Transcripts are 100% on-device by default using its bundled Whisper Pro and Parakeet V3 models; recordings stay encrypted on your Mac. It auto-detects when a meeting starts, which removes the 'forgot to hit record' failure mode.

Pricing as of June 2026: a free plan that works with your own API keys or local models; Local+ at $29/year, which unlocks Alter's real-time local speech-to-text models on up to 3 devices; Pro at $240/year for unlimited cloud-model requests and unlimited meeting recordings with speaker identification; and a $720 lifetime license. The $29/year Local+ tier is the cheapest commercial on-device transcription on this list.

Best for: Mac users who want local meeting transcription plus a broader AI assistant. Honest limitations: macOS only; the full experience (unlimited recordings, speaker ID, cloud models) sits in the $240/year Pro tier, and using its cloud AI features reintroduces exactly the data flow you may be trying to avoid — keep to Local+ if that matters.

6. Granola — bot-free is not the same as cloud-free (the caveat entry)

Granola deserves its popularity: no bot joins your call, audio is captured directly from your device, and its AI notes are widely considered excellent. It is frequently recommended as the 'private' Otter alternative, and for the social problem — a recording bot staring at your client — it genuinely is. But Granola's transcription and note generation run on cloud servers. Your audio-derived data leaves your machine; you are trusting Granola's policies (it offers model-training opt-out, with org-wide enforcement on Enterprise) rather than your own hardware. Its pricing page does not describe on-device processing, because that is not the architecture.

Pricing as of June 2026, per Granola's official pricing page: a free Basic plan with AI notes and limited meeting history; Business at $14/user/month with unlimited history, integrations (Notion, Slack, HubSpot, Zapier), and API access; Enterprise at $35/user/month with SSO and admin controls.

Best for: teams who want invisible capture, polished collaborative notes, and integrations, and whose compliance posture permits cloud processing. If your requirement is 'audio never leaves the device,' Granola does not meet it — every other tool on this list does.

7. Apple's built-in call recording and transcription (iOS 26)

If what you actually need to transcribe is phone calls, your iPhone may already do it. iOS 26's Phone app records calls (announcing the recording to all participants) and generates transcripts on-device — Apple's documentation states transcripts are produced locally rather than on servers. With Apple Intelligence enabled, you also get on-device call summaries. The Notes app offers similar on-device audio transcription.

Price: free, built into iOS 26. Significant restrictions apply as of June 2026: call recording is unavailable in the European Union and a number of other countries (including Turkey, Saudi Arabia, the UAE, Russia, and South Africa), and transcription requires the device language to be set to one of about ten supported languages, including English, Spanish, Portuguese, French, German, Japanese, Korean, and Chinese.

Best for: iPhone users in supported regions who need occasional call records at zero cost. Honest limitations: it covers cellular/FaceTime calls only — not Zoom, Meet, Teams, or in-person meetings — and offers no speaker identification across meeting apps, no meeting library, no export workflow, and no cross-device search.

How to verify a tool is actually on-device

Vendors say 'private' loosely, so verify instead of trusting copy. For a web app, open your browser's DevTools (F12), switch to the Network tab, start recording, and talk for a minute: an on-device tool shows no outbound requests carrying audio payloads — model files download once, nothing uploads. For desktop apps, the blunt test is disconnecting from the internet mid-meeting; a genuinely local tool (Meetily, MacWhisper, Superwhisper, Alter's local models, now recording after models are cached) keeps transcribing. A network monitor like Little Snitch on macOS makes the same check continuous.

Also read what the exceptions are. Most local-first tools, including ours, send transcript text to a cloud LLM only when you explicitly request an AI summary — that is a different, smaller exposure than uploading raw audio, but it is not nothing, and a vendor that discloses it plainly is telling you they have thought about the boundary.

frequently asked questions

Does Otter.ai upload and store my audio in the cloud?

Yes. Otter.ai's architecture is cloud-based on every plan: a bot joins your meeting (or you import/record audio), and the recording is uploaded to and stored on Otter's servers, where transcription happens. As of June 2026, the free plan includes 300 transcription minutes per month and Pro costs $16.99/month billed monthly or $99.99/year. If your requirement is that audio never leaves your device, Otter cannot meet it regardless of plan.

Is on-device transcription as accurate as Otter's cloud transcription?

It is close, and for English and major languages often indistinguishable. Most tools on this list run Whisper-class models — the same family of open models many cloud services build on. Accuracy depends on which model size your hardware can run: an Apple Silicon Mac or a WebGPU-capable browser handles models that rival cloud output, and several tools (including now recording's polish pass and MacWhisper's larger models) offer a second high-accuracy pass. Very old hardware limits you to smaller, less accurate models.

How can I check that an app really isn't uploading my recordings?

Two practical tests. For web apps: open DevTools, watch the Network tab while recording — you should see model files download once and no audio leave the machine. For desktop apps: disconnect from the internet mid-recording; a genuinely local tool keeps transcribing. A firewall monitor like Little Snitch (macOS) verifies this continuously. Marketing language like 'private' or 'bot-free' is not evidence — Granola is bot-free and still processes audio in the cloud.

Do these tools work with Zoom, Google Meet, and Microsoft Teams?

Yes, with one exception. Because local-capture tools record your device's microphone and system audio rather than joining the call, they work with any meeting software — Zoom, Meet, Teams, Webex, a phone on speaker, or an in-person conversation — with nothing to integrate and no bot appearing in the participant list. The exception is Apple's iOS 26 feature, which records cellular and FaceTime calls only.

If no bot joins the call, do I still need consent to record?

Yes. Recording consent laws apply to the act of recording, not the method. In the US, some states require all-party consent (California, Florida, Illinois, among others) while federal law and most states require one-party consent; many other countries have stricter rules. A bot at least announces itself — with silent local capture, the disclosure burden is entirely on you. Tell participants you're recording. Apple's iOS 26 feature plays an announcement automatically for this reason.

What's the catch with tools that don't upload audio?

Three honest trade-offs. First, hardware: transcription runs on your machine, so quality and speed scale with your CPU/GPU. Second, features: cloud incumbents are ahead on video recording, CRM and calendar integrations, and team admin controls. Third, AI summaries: generating good summaries locally requires a powerful machine (Meetily via Ollama), so most tools — including now recording — send transcript text (never audio) to a cloud LLM for that one optional feature, and you should confirm any tool's exceptions before trusting it.