Most AI Meeting Tools Store Your Transcripts as Plain Text. I Built One That Doesn't.
AI meeting assistants have become incredibly good over the past year.
They can transcribe conversations, answer questions during meetings, summarize action items, and even coach you during interviews.
But while trying a bunch of them, I kept wondering about one question that surprisingly few products talk about:
Who can actually read my meeting history?
For most AI meeting tools, transcripts, notes, titles, and chat history are stored as ordinary plaintext in a database. They're encrypted while being transmitted and while sitting on disk, but once the application needs to read them, they're decrypted on the server.
That's a perfectly common architecture.
It just wasn't one I wanted for my own meetings.
So while building Observer, I decided to solve a different problem first:
What if the server never stored readable meeting history in the first place?
The problem with meeting AI
Modern meeting assistants are becoming your second brain.
They remember:
- interviews
- customer calls
- salary discussions
- design reviews
- roadmap planning
- legal conversations
That's a huge amount of sensitive information.
Yet most of us barely think about where all of it ends up.
After using several products, I realized I cared less about whether AI could summarize my meeting...
...and more about who could browse those summaries later.
Building Private Vault
Observer stores meeting history.
It has to.
Otherwise features like searching previous meetings or asking follow-up questions wouldn't work.
The difference is how it's stored.
Before meeting data is written to the database, it's encrypted on your Mac using AES-256-GCM.
That includes:
- meeting titles
- transcripts
- meeting notes
- AI conversations
The encryption key comes from your passphrase.
That passphrase never leaves your device.
The server stores ciphertext—not readable meeting content.
So instead of saying
"Trust us, we won't look."
the architecture is closer to
"We couldn't casually look even if we wanted to."
Of course, there are practical limitations.
Some metadata still needs to remain accessible so the product can work:
- account email
- subscription status
- meeting duration
- usage statistics
And whenever AI needs to answer a question or generate notes, the meeting is temporarily decrypted, processed, and immediately encrypted again.
I think that's a much more honest privacy story than claiming everything is magically "end-to-end encrypted."
AI without another meeting participant
The second thing that bothered me was meeting bots.
Most assistants join Zoom or Meet as another attendee.
Someone inevitably asks:
"Who's that?"
Observer doesn't join the meeting.
It listens through your Mac's system audio, so it works with:
- Zoom
- Google Meet
- Teams
- Slack
- Discord
- Webex
- browser-based meetings
No participant.
No calendar integration.
No special setup.
Live assistance instead of post-meeting summaries
Most products shine after the meeting is over.
I wanted help while I was still talking.
Observer supports different actions depending on what you need:
- Answer → generate a quick reply from the conversation
- Assist → combine transcript with what's on your screen
- Recap → summarize the last few minutes
- Screen Analysis → explain whatever is currently visible
- Chat → ask follow-up questions naturally
The live transcript becomes the AI's working memory instead of disappearing into a summary afterward.
One feature I didn't expect to love
After every meeting, Observer stores:
- transcript
- notes
- AI conversations
The feature I use the most isn't actually AI.
It's clicking a note and jumping directly to the exact moment in the transcript where that topic was discussed.
Instead of wondering:
"Didn't we decide this two weeks ago?"
I can immediately find the original conversation.
Small feature.
Huge quality-of-life improvement.
Notion support
I didn't want another knowledge base.
Most of my documentation already lives in Notion.
Observer simply lets me export meeting notes into an existing Notion page or database whenever I want.
Meeting history stays encrypted inside Observer.
Notion remains my long-term workspace.
Observer vs. Cluely
People naturally compare Observer with Cluely, and I think they're solving similar—but not identical—problems.
Cluely has built a much larger brand and offers a polished AI overlay experience across multiple platforms.
Observer focuses on a slightly different set of priorities.
| Feature | Observer | Cluely |
|---|---|---|
| Meeting bot required | No | No |
| Device-side encrypted meeting history | ✅ Private Vault | Focuses on encryption in transit and at rest |
| Live transcript | Always visible | Available, but less central to the workflow |
| Notion export | Built in | Public integrations focus more on CRM/ATS |
| AI workflow | Separate Answer, Assist and Recap actions | More unified assistant experience |
| Undetectability | Included in paid plans | Higher-tier plans emphasize this capability |
| Platforms | macOS today, Windows coming | macOS and Windows |
Neither approach is objectively better.
If you're looking for a mature AI assistant with a broader ecosystem, Cluely is a strong choice.
If you're someone who cares about where meeting history lives, who can read it, and wants encryption to be part of the product architecture—not just a security page—then Observer may be worth a look.
Why I built it
I didn't build Observer because the world needed another AI meeting assistant.
I built it because I wanted one that I would actually trust with my own conversations.
AI is getting incredibly good.
The next differentiator won't just be model quality.
It'll be how responsibly we handle the information users trust us with.
I'd love to hear how others are approaching this.
If you're building AI products:
How are you thinking about user data and privacy?
Top comments (0)