Most AI meeting tools work roughly the same way: they join your call, record it, send the audio file to their cloud servers, and email you a summary an hour later. That flow is convenient right up until someone asks where those audio files are stored, how long they're kept, and who can access them. The answer, in most cases, is "read the terms of service."
Meetily takes a different position. It runs the transcription engine directly on your machine — nothing leaves your computer. The same AI that powers the transcript runs locally, so even the model computation stays inside your walls. For a lawyer discussing client matters, a recruiter handling confidential compensation conversations, or an analyst sitting in on a sensitive board briefing, that distinction is not minor.
What Meetily actually is
Meetily is a desktop application (macOS and Windows) that captures your system audio and microphone simultaneously, runs a speech-to-text model directly on your device, and generates meeting summaries through an AI provider of your choice — including a fully local option called Ollama that requires no API key and no external calls whatsoever.
The transcription engine uses either Whisper or Parakeet — two open-source speech recognition models. Whisper is a widely used model from OpenAI (open-sourced in 2022); Parakeet is a newer model from NVIDIA that tends to run significantly faster on modern hardware. The project claims up to four times faster transcription than typical Whisper implementations, primarily because of hardware acceleration: Apple Silicon uses Metal and CoreML, while Windows machines can take advantage of NVIDIA's CUDA or AMD/Intel GPU acceleration via Vulkan. These terms just mean the app uses your graphics chip for the heavy computation, which is where those speed gains come from.
The application is built on Tauri, a framework for building desktop applications with a Rust backend and a web-based frontend. In practice, it means the app is a single self-contained download — no runtime dependencies, no Python environment to manage, no command line required.
| Fact | Value |
|---|---|
| What it is | Local AI meeting assistant (desktop app) |
| Built by | Zackriya Solutions |
| License | MIT — free to use and modify |
| Platforms | macOS, Windows (Linux via source build) |
| Requires | System audio access; optional AI API key for summaries |
| Install method | Download installer from GitHub Releases |
| Verdict | Install if your meetings are confidential; evaluate if you just want quick notes |
What you get after installing it
Open the app and it immediately presents an audio device picker — microphone and system output, captured simultaneously. From there, you click Record and talk. A live transcript appears in real time on the right side of the screen as you speak. The entire pipeline — audio capture, on-device AI processing, transcript rendering — runs locally without any network calls.
When the meeting ends, you hit Stop and then request a summary. This is the one step where you have a choice: you can run the summary through a local Ollama instance (free, no API key, no data leaves the machine), or connect an API key for Claude, Groq, OpenRouter, or any OpenAI-compatible service (any AI provider that speaks the same API format as OpenAI) if you prefer the quality of a hosted model and are comfortable with that data policy.
The summary output covers the expected ground: key points, decisions, and action items. It is not deeply configurable in the community edition — you get what the prompt produces, not a template system. Meetily Pro adds custom summary templates for users who need specific formats (legal, engineering sprint, board meeting, and so on).
Beyond live recording, the app also supports importing existing audio files — if you have a recording from a previous call in MP4 or WAV format, you can drop it in and generate a transcript. The feature is marked beta, but it works. You can also re-transcribe any existing recording with a different model, which is useful if you recorded a meeting when only a smaller model was installed and later want higher accuracy.
Everything the app stores — transcripts, recordings, model files — lives on your local disk. There is no account required. There is no sync to any cloud service. The data model is file-based and accessible.
The install experience
Installation is a download-and-run process. Navigate to the Releases page on GitHub, download the appropriate file (the dmg installer for Apple Silicon Macs, the exe file for Windows), and run it. On macOS, you drag the app to your Applications folder in the standard way.
First launch triggers a one-time model download. Whisper or Parakeet needs to be pulled from the internet before local transcription starts working — the model files are several hundred megabytes, so this step takes a few minutes depending on your connection. After that, everything runs offline if you choose the Ollama path for summaries.
The audio permission prompts are the other thing worth flagging. Capturing system audio on macOS requires a system permission that Apple treats as sensitive; the app walks you through this, but it involves a trip to System Settings and a restart of the app. It is not complicated, but it is a few steps, and first-time users who have never granted system audio permissions to an app may need to follow the in-app instructions carefully.
The Linux path requires building from source — cloning the repository and running a build script. It works, but it requires some comfort with a terminal and is meaningfully more involved than the macOS and Windows experience. Developers will find it routine; non-technical Linux users should proceed with some patience.
Where it fits and what to compare it to
The obvious comparisons are the cloud meeting tools that dominate this space: Otter.ai and Fireflies.ai. Both are polished consumer products with calendar integration, speaker diarization (identifying who said what), meeting bot joining (the bot joins your Zoom call without you doing anything), search across all past transcripts, and team sharing. Their transcription accuracy on clear audio is excellent.
The tradeoff is that both upload your audio to their servers. Their privacy policies explain what is stored and for how long, and both have enterprise tiers with stronger data handling commitments. But the audio file does leave your device. For conversations that are genuinely sensitive — legal work, deal negotiations, personnel discussions, health-related conversations, government contracts — that is a policy question that may not have a comfortable answer.
Meetily's position in that comparison is unambiguous: if the requirement is "audio never leaves this device," Meetily satisfies it and the cloud tools do not. If the requirement is "I want the most convenient, richest-featured meeting tool with calendar integration and a team dashboard," the cloud tools win easily.
The honest caveats: speaker diarization (the feature that labels transcript lines by speaker) is not in the community edition — it is listed as a Meetily Pro feature. The summary formatting is fixed. There is no calendar integration, no meeting bot that joins calls autonomously, and no searchable history across all past meetings in a web interface. For professionals who need those features and have cleared the data handling question internally, the cloud tools remain the practical choice.
For the specific persona who cannot or will not send audio offsite, Meetily is currently the most usable local option — more so than building something similar from raw Whisper components, which would require considerably more technical setup.
Verdict
Meetily works, and it works for the right reason: the privacy guarantee is structural, not a policy statement. The processing happens locally, so there is no policy to trust — there is no external service involved in the transcription step at all.
The gaps are real. No speaker diarization in the free version, no meeting bot, no calendar sync, limited summary customization. If those features are blockers, Meetily Pro or one of the cloud tools is the answer.
If your situation is "I handle confidential conversations and the idea of sending those recordings to a third-party server is not acceptable," Meetily is the practical starting point. The install is ten minutes, the model runs on your machine, and the transcript is yours.
For your agent to install this
Your AI agent (Claude Code, Cowork, ChatGPT Work, Codex, or Hermes) can install Meetily without any manual steps using the instructions below.
macOS (Apple Silicon):
curl -L -o ~/Downloads/meetily.dmg \
"https://github.com/Zackriya-Solutions/meeting-minutes/releases/latest/download/meetily_0.4.0_aarch64.dmg" \
&& open ~/Downloads/meetily.dmg
After the disk image opens, drag Meetily to /Applications, then open it from the Applications folder. Grant microphone and system audio permissions when prompted and download the transcription model on first launch.
Windows (x64):
curl -L -o %TEMP%\meetily-setup.exe ^
"https://github.com/Zackriya-Solutions/meeting-minutes/releases/latest/download/x64-setup.exe" ^
&& start %TEMP%\meetily-setup.exe
Run the installer and follow the setup steps. For the latest release file names, check the Releases page directly — file names include the version number and may change between releases.