Your standup notes should not live on someone else's GPU next to a thousand other companies' board decks. Yet most meeting bots want microphone access, calendar hooks, and a permanent seat in the cloud. Meetily flips that: a Tauri desktop app (Rust backend, Next.js UI) that records, transcribes, and summarizes on hardware you control, with Whisper or NVIDIA's Parakeet models running locally and Ollama as the default summarizer.
It is MIT licensed, heavily starred, and aimed at people who need real meeting minutes without shipping audio to a SaaS transcription API. The community edition is free; the vendor also sells Meetily PRO for teams that want managed accuracy upgrades and (eventually) richer diarization.
What it actually does
During a call, Meetily captures mic plus system audio (with mixing that tries not to clip when you and Zoom talk at once). It streams speech to text on the machine using Whisper or the faster Parakeet path the project advertises. When the meeting ends, you pick an AI backend for the summary: Ollama locally is the recommended default, but Claude, Groq, OpenRouter, or your own OpenAI-compatible endpoint work too.
You can import existing audio files and re-run transcription in another language or model (beta). Everything lands in local storage: recordings, transcripts, settings. No account required for the open-source build.
Platform support today is practical, not universal: signed installers for Windows and macOS, and Linux via source build (pnpm + their GPU build scripts). Architecture docs describe a single self-contained Tauri app, not a classic LAMP stack in a container.
Why "self-hosted" here means something specific
Meetily is not Nextcloud in a Docker one-liner. For most users, self-hosting means your laptop or workstation owns the pipeline: models on disk, transcripts in your home folder, summaries via local Ollama. That is the privacy win for lawyers, clinics, defense contractors, and anyone who has been told "no external AI" by compliance.
If you already run Ollama on a Canadian VPS, you can point Meetily's summary step at that endpoint (OpenAI-compatible URL) so inference stays on infrastructure you operate, even while capture happens on a staff MacBook. Splitting capture (desktop) and inference (private server) is a common pattern for small teams that cannot put GPU on every laptop.
Meetily PRO advertises fuller team self-hosted deployment for orgs that outgrow one-off installs. Community edition remains the hackable baseline; PRO is the commercial lane with templates, exports, and roadmap items like auto-join and speaker ID.
Who it is for (and who should pass)
Good fit: consultants and agencies who record client calls under NDA, remote teams that want Otter-style workflows without Otter's cloud, and Canadian orgs documenting PIPEDA-friendly processing (audio stays in Canada if the machine stays in Canada).
Skip or plan carefully if: you need a pure headless server that joins Google Meet without a user desktop (community Meetily is not that yet), you are on older CPUs without AVX2 (Windows builds call this out), or you expect turnkey Linux packages. Budget time to build from source on Linux or standardize on macOS/Windows clients.
Hardware and hosting notes
Transcription speed depends on the build: Metal on macOS packages, Vulkan on Windows Whisper builds, optional CUDA if you compile for NVIDIA. Parakeet is the "faster" path when you have the right stack. Summaries via Ollama need RAM for the model you choose; a 7B class model on the same machine as live transcription can feel tight on 16 GiB laptops.
For a team hub, a Docker-friendly VPS in Beauharnois or Toronto running only Ollama (or vLLM behind an internal URL) can serve many desktops without exposing models to the public internet. Lock it to VPN or private network, TLS, and auth on the API path. Meetily still records locally; the server only sees text prompts for summarization if you configure it that way.
Compliance reality check
Local processing removes a whole class of vendor risk, but it does not magically make recording legal. You still need consent where your jurisdiction requires it, retention policies for stored audio, and disk encryption on laptops that leave the office. Meetily gives you control; your policy team still writes the rules.
Bottom line
Meetily is one of the more polished open-source answers to "meeting AI without the cloud." It is desktop-first, Rust-serious, and honest about PRO upsells. For Canadian teams, pairing local capture with Ollama on your own metal is a credible architecture that auditors can follow.
Want a private Ollama endpoint on Canadian infrastructure for tools like Meetily? Talk with Swift Host. We will size CPU, RAM, and disk for the models you actually run, not the brochure specs.
