AI Meeting Summaries Without Sending Your Audio: Local or Bring-Your-Own-Key

Most AI meeting tools make you choose between two bad deals.

Cloud note-takers are convenient, but they want everything. A bot joins your call, your audio is uploaded to their servers, and you summarize one of the most sensitive recordings in your organization by trusting someone else's privacy policy. Strategy discussions, personnel decisions, client negotiations - all of it leaves the building.

The other camp is local-first tools that keep your data private but ask you to assemble the AI pipeline yourself. That was real friction, and we heard it from users constantly.

"Wait, I need to get an API key from OpenAI?"

"What's Ollama? Do I need to install something else?"

"I just want meeting notes. Why does this involve so much setup?"

Fair questions. So we want to be clear about what Meetily actually does, and what your real options are when it comes to summaries - because you do not have to upload your audio to get good ones.

The Part Nobody Should Have to Compromise On

Here is the line we will not move: your audio stays on your device.

Meetily captures your meeting and transcribes it locally with Whisper. That step is always on-device, on every tier, with no exceptions. No bot joins the call, and the recording never gets uploaded for transcription.

Summaries are the step where most tools quietly send everything to the cloud. We split that step out and made it pluggable, so you control where (and whether) any text goes.

Two Ways to Summarize Without Uploading Audio

For summary generation, you have two privacy-respecting paths. Both work without our keys, and neither one sends your audio anywhere.

Option 1: Local Model (Fully On-Device)

Run the summary on your own machine through a local model via Ollama, or the built-in option. Nothing leaves your device - not the audio, not even the transcript. This is the right choice for regulated industries, air-gapped environments, or anyone who wants the entire pipeline to stay local. It asks more of your hardware, and smaller local models won't match a frontier model on the hardest summaries, but for everyday meetings they do the job and keep everything in one place.

Option 2: Bring Your Own Key (BYOK)

When you want a frontier model - Claude, OpenAI's GPT, Gemini, Mistral, Groq, OpenRouter, or Azure OpenAI - connect your own API key. You control the provider, the model, and the billing through your own account. Here only the text transcript is sent to the provider you chose. Your audio still never leaves your device. You get top-tier summary quality without handing your recordings to a meeting-bot vendor.

Why Not Just Upload Everything?

You might wonder why we didn't build the simple version: upload audio, get notes back. Plenty of tools do exactly that.

Because of what meeting recordings actually contain. Strategy, headcount, client terms, product roadmaps, the offhand comment someone would never put in writing. Sending all of that to a third-party server, even an encrypted one, is a trade many teams aren't willing to make - and once it's uploaded, you're trusting a policy, not an architecture.

Our approach draws the line differently:

• Audio stays on your device. Always. No exceptions.
• Transcription runs locally. Whisper processes your audio right on your machine.
• Only text leaves your device - and only if you choose BYOK for the summary.
• You pick the trade-off. Everything on-device with a local model, or frontier-quality summaries with BYOK where just the transcript travels.

This is what we mean by "privacy-first with pluggable AI." You're never locked into one approach, and the most sensitive part of the recording stays put either way.

Which Option Should You Use?

You can switch between these at any time. Use a local model for that sensitive board meeting, then switch to BYOK for routine standups where you want the sharpest possible notes. The architecture supports mixing and matching because one size doesn't fit all - and unlike a cloud note-taker, your audio is off-limits no matter which you pick.

How This Compares to Cloud Note-Takers

Cloud-based assistants like Otter.ai and Fireflies handle everything for you, and that's the point and the problem. To get a summary, your audio goes up to their servers, and a bot usually has to join the call to capture it.

Meetily inverts that. There is no bot. The audio is transcribed where it was recorded. And the summary either stays fully on-device (local model) or sends only the transcript to a provider you chose and pay (BYOK). You're not trusting us with your recordings, because we never receive them.

What We Learned Building This

A few things became clear while building Meetily's summary pipeline:

1. Privacy isn't binary. Users don't want "all local" or "all cloud." They want control. Some meetings need everything to stay on the machine. Others just need sharp notes and a transcript can travel. A good AI meeting notes tool should support both.

2. Setup was a real barrier. People abandoned the summary step more than any other. The answer wasn't to upload their audio for them - it was to make the on-device and BYOK paths clear, so the privacy-respecting option is also the easy one.

3. Transcription and summarization have different privacy needs. Audio is uniquely sensitive - it carries voice biometrics, background conversations, ambient sound. A text transcript is sensitive too, but less so. Splitting these into separate steps, with separate privacy choices, is the whole design.

4. Trust is earned through architecture, not promises. We could have shipped the simple product that uploads everything and asks you to trust the policy. Instead we kept audio on the device and let you choose how summaries are generated. That's slower to build and easier to trust.

How to Try It

If you're new to Meetily:

1. Download Meetily from meetily.ai(opens in new tab)
2. Record a meeting and get a local transcript automatically
3. Generate a summary with a local model (fully on-device) or your own API key
4. Switch the engine any time, per meeting, based on how private it needs to be

No bots join your calls. No audio leaves your device. Just meeting notes that respect where your data lives.