Your AI Notetaker Is a Sub-Processor. Meetily Removes It From Client Calls

Every agency and professional services firm has quietly added an AI notetaker to client calls in the last two years. Almost none of them have told the client, and fewer still could name the vendor that now holds a transcript of every strategy conversation. A free, open-source, entirely local alternative changes that math — but not for the reason the headline suggests.

“There are endless ways to record and transcribe your virtual meetings with AI. Here’s an option that’s free and open source.”

— Justin Pot, WIRED

Our take

Meetily is an MIT-licensed meeting assistant that records, transcribes and summarizes on your own machine. It ships as an installer for macOS and Windows, builds from source on Linux, and runs transcription through either OpenAI’s Whisper or NVIDIA’s Parakeet model, with GPU acceleration on Apple Silicon, CUDA and Vulkan. Summaries can run through a local model via Ollama or, if you choose, a hosted one. All of that is documented in the project repository.

The subscription savings are real and also the smaller half of the story. Published business-tier pricing runs about $20 per user per month for Otter, $19 for Fireflies, $14 for Granola and roughly $19 for Fathom (comparison here). A six-person team on Otter Business is roughly $1,440 a year. That is a line item worth cutting, but it is not the reason to do this.

The reason is that a cloud notetaker is a sub-processor sitting inside your client relationships. When you run Otter or Fireflies on a client strategy call, a third company now holds the recording, the transcript and a searchable index of everything that was said. If you have signed a data processing agreement, that vendor belongs on your sub-processor list. If you serve healthcare, legal or financial clients, it may need explicit consent. Most small agencies have never done either. Processing locally deletes the entire question, and that is worth far more than $1,440.

Now the second-order effect nobody is discussing. Cloud notetakers usually join the call as a visible participant — a bot with a name that appears in the roster. That visibility is annoying, and it is also doing legal work for you: it puts everyone on notice that recording is happening. A local tool capturing system audio announces nothing. Convenient, and genuinely risky.

Recording consent law is not uniform and it is not intuitive. Nevada is a hybrid state: one-party consent for in-person conversations, but all-party consent for telephone and wire communications under NRS 200.620, following the state Supreme Court’s decision in Lane v. Allstate (summary here). California requires all-party consent. Your clients are not all in your state. Silent local capture removes the one cue that used to prompt the disclosure conversation, so you have to build that disclosure into your own call script instead. This is general information, not legal advice — ask your attorney about your specific setup.

Then there is the setup detail that will bite people hardest, and the source treatment glides past it. Meetily lets you point summarization at Ollama running locally, or at Claude, Groq, OpenRouter or any OpenAI-compatible endpoint. If you pick a hosted backend for convenience, you have reintroduced exactly the third-party processor you were trying to eliminate. The transcription stays local; the summary does not. Half-local is not local. If confidentiality is the point, run Ollama and verify it.

We also want to push back on “free.” The software costs nothing. The deployment does not. Whisper plus a local language model wants real hardware — a recent Apple Silicon Mac or a machine with a capable GPU and enough RAM to hold a model in memory. On a five-year-old office desktop, transcription will be slow and summarization may not run at all. The honest comparison is not $0 versus $20 per user. It is $0 in software plus possibly a $1,500 to $2,500 machine and a few hours of setup per seat, versus $20 per user with no setup and a support number.

Which leads to our actual recommendation, and it is not the one the enthusiasm suggests. Do not roll this out across your whole team. Open source means you own the updates, the model downloads and the troubleshooting, and there is no vendor to call when a recording fails ten minutes into a new-business pitch. Run it on the two or three machines where confidentiality genuinely matters — client strategy sessions, HR conversations, anything under an NDA — and keep the paid cloud tool for internal standups where nobody cares. Mixed stacks are fine. Pretending one tool fits every conversation is what got everyone into this in the first place.

What this means for your business

1. Pull your sub-processor list and check whether your notetaker is on it. If you have signed any DPA and are running Otter, Fireflies or Fathom on client calls, that vendor almost certainly belongs on the list. This is a one-hour audit that prevents a very bad conversation later.
2. Pilot on one machine, not the whole team. Install Meetily on the laptop that runs your most sensitive calls. Test it on three real meetings before you cancel a single subscription. Transcription accuracy on accented speech and crosstalk is where local tools usually disappoint.
3. If you use it, force the local summarization backend. Configure Ollama and confirm no hosted API key is set. Choosing Claude or Groq for summaries undoes the entire privacy rationale while making you feel like you solved it.
4. Rewrite your call opening. Because local capture is silent, add one sentence to your standard meeting intro announcing that you are recording and asking for agreement. Ten seconds, and it covers you in all-party consent states where your clients may live.
5. Budget the hardware honestly. Price the machine before you price the savings. If your team is on aging Windows laptops, the cloud subscription is probably still the cheaper path for another refresh cycle.

The real win here is not saving a subscription. It is being able to tell a client, accurately, that nothing said on your calls leaves your building. Read the full setup walkthrough at WIRED.

---