Local-first
Local-first meeting intelligence for Mac
Not “private by default.” Not “encrypted in transit.” Local-first means the audio never leaves your Mac.
Most transcription tools use “privacy” as a label. Thunder Kitty uses it as an architecture. Real-time transcription, on-device AI, and meeting intelligence — processed on your Mac, stored on your Mac, never sent anywhere.
What local-first means
Local-first isn't a marketing term. It's a technical architecture. Audio is processed on your device. Transcripts are stored on your device. AI runs on your device. Nothing is routed through a server — not ours, not a partner's, not anyone's.
Most transcription tools call themselves private, then send your audio to Deepgram, AssemblyAI, or OpenAI's Whisper API for processing. The audio leaves your machine. It travels over the internet. It lands on someone else's infrastructure. That's not local-first. That's cloud-based with a privacy label. Those are companies. They have employees. They have data retention policies. They have lawyers. Your audio is in their systems, under their terms.
Local-first means the audio never leaves at all. Not temporarily. Not encrypted. Not “anonymized.” It stays on your Mac because it's processed on your Mac.
Verify it yourself: turn on Airplane Mode. Thunder Kitty still works. That's not a demo trick — it's the architecture.
What we're building toward
Thunder Kitty's goal is to push the boundary of what's possible in local-first meeting intelligence on macOS. Today that means real-time transcription using Apple's SpeechAnalyzer, on-device AI summaries via Apple Intelligence and local models, and a notepad-first experience designed around how professionals actually think before, during, and after a meeting.
We're building real-time meeting intelligence that runs entirely on your Mac — a live topic timeline that shows the structure of your meeting as it unfolds, agenda tracking that tells you which items you've covered and which you haven't, and per-agenda-item AI summaries generated after the meeting using a local model. These aren't cloud features waiting to be ported on-device. They're capabilities that are only practical locally — continuous processing, zero marginal cost, zero latency — things that would cost dollars per meeting and add noticeable delay if they ran in the cloud.
The pace of on-device AI capability is accelerating. We're building at the front of it, not waiting for the cloud to catch up.
Why macOS
Apple Silicon and the Neural Engine make on-device AI genuinely fast. Apple's SpeechAnalyzer — available in macOS 26 and later — is a pro-grade speech recognition API that runs entirely on-device. It's the same engine Apple uses for Apple Intelligence.
Other tools use Whisper.cpp or cloud APIs. Thunder Kitty uses the platform's native engine. That's not a positioning choice — it's a technical one. SpeechAnalyzer is faster, more accurate, and more power-efficient than anything we could bundle ourselves.
We build for the platform that makes local-first possible at the quality level professionals expect. Right now, that platform is macOS.
What you can verify
We'd rather you test the claims than trust the copy.
- Turn on Airplane Mode. Transcription still works.
- Open Activity Monitor → Network while recording. You'll see nothing.
- Block Thunder Kitty in your firewall. It still works.
- Your notes are plain Markdown files in your Documents folder. Open them in any app.
Who this is for
Therapists whose clients trust that what they say stays in the room. Lawyers who can't introduce a third party into a privileged conversation. Consultants whose engagement letters promise confidentiality they need to actually deliver. Founders in fundraising conversations. Executives in board meetings. Sales reps on calls where the prospect's words matter. Anyone who handles conversations too sensitive for the cloud.
Or anyone who simply believes their notes shouldn't live on someone else's server.
Common questions
What’s the difference between ‘local-first’ and ‘end-to-end encrypted’?
End-to-end encryption means data is encrypted before it leaves your device and decrypted at its destination — but it still travels to someone else’s servers. The data exists on external infrastructure, encrypted. Local-first means the data never leaves your device at all. Not encrypted-then-transmitted. Not stored somewhere encrypted. Just: never sent. The distinction matters when the question is whether any third party ever has access to your conversations — even theoretically, even encrypted.
How is ‘local-first’ different from ‘privacy-first’ as a marketing claim?
Privacy-first describes a policy — a company’s stated intentions about how they’ll handle your data. Local-first describes an architecture — your data physically cannot reach a third party because it’s processed on your device. One is a promise that depends on a company keeping it. The other is a technical constraint that doesn’t require trust.
Can I actually verify that nothing leaves my Mac?
Yes — in several ways. Turn on Airplane Mode and record a meeting: transcription still works. Open Activity Monitor → Network tab while recording: you’ll see no Thunder Kitty network activity. Block Thunder Kitty in your Mac’s firewall: it still works. Your notes are plain Markdown files in your Documents folder: open them in any text editor. The claims are testable. We’d rather you test them than take our word for it.
If Thunder Kitty goes away as a company, what happens to my data?
Nothing changes. Your notes and transcripts are plain Markdown files on your Mac. You lose the recording and transcription functionality. You don’t lose your data — it was never ours. Open your notes in Obsidian, iA Writer, VS Code, or any text editor. They’re files, and files outlast any app.
Does local-first mean Thunder Kitty can never add team features?
Not necessarily. Local-first describes where data lives, not what features are possible. Team features built on shared local files — a shared Dropbox or iCloud Drive folder full of Markdown files, visible to everyone on the team — are entirely compatible with local-first principles. Team features that require routing audio or transcripts through a cloud service are not. The constraint is architectural, not a product philosophy against collaboration.
Is there a quality trade-off for going local-first?
There was, and it’s narrowing fast. Apple’s SpeechAnalyzer on macOS 26 is genuinely excellent for transcription — fast, accurate, and it runs on-device with zero latency. For AI summaries, on-device models are strong for structured use cases like per-agenda-item summaries, and improving with every new model release. If you want more powerful inference, Thunder Kitty lets you connect your own Claude, OpenAI, or Gemini API key — your data goes to your account under your terms. The local-first architecture doesn’t force a quality ceiling. It forces a choice about what trade-offs you’re making consciously.
Try the most private meeting tool on macOS.
7-day free trial. No credit card. No account. Just a download.
Download for macOS