Meetily

Meetily is a self-contained desktop application for professionals and organizations who need to record, transcribe, and summarize meetings without trusting any cloud service. Every step of the pipeline — audio capture, speech recognition, and AI summarization — runs on your own machine using open-source models. This makes it genuinely useful for compliance-sensitive contexts like legal, healthcare, defense, and enterprise teams operating under strict data residency requirements.

The application ships as a single Tauri binary combining a Rust backend and a Next.js frontend. For transcription it offers two swappable engines: whisper.cpp for broad compatibility and Parakeet (via ONNX Runtime) for significantly faster throughput on modern hardware. GPU acceleration is compiled in automatically at build time for Metal and CoreML on Apple Silicon, and for CUDA, Vulkan, and HIP on Windows and Linux. For summarization it supports local Ollama inference, a bundled llama.cpp sidecar for fully offline operation, and optional cloud providers like Claude, Groq, or any custom OpenAI-compatible endpoint.

Meetily stores all recordings, transcripts, and summaries locally in a SQLite database with automatic migrations. The audio engine captures microphone and system audio simultaneously with FFmpeg-based mixing, Voice Activity Detection, and intelligent level management. An import feature lets users drop in existing audio files in any common format for local re-transcription. Pause/resume, system tray controls, OTA updates, and a BlockNote-based rich text editor for editing summaries round out the desktop experience.

A Meetily PRO tier is available separately at meetily.ai/pro for teams needing enhanced transcription accuracy, speaker diarization, advanced export formats, and calendar integration. The open-source Community Edition remains free and covers the full local transcription and summarization workflow indefinitely.

What You Get

• Dual Transcription Engine - Choose between Whisper.cpp for broad model compatibility or the faster Parakeet ONNX-based engine, both running entirely offline with no audio ever sent to external servers.
• Fully Offline AI Summaries - Summarize meetings using a bundled llama.cpp sidecar or your local Ollama instance, with optional support for Claude, Groq, OpenRouter, and custom OpenAI-compatible endpoints.
• Simultaneous Mic and System Audio Capture - Captures microphone and system audio in parallel with FFmpeg mixing, Voice Activity Detection, and intelligent RMS-level management to prevent clipping.
• GPU Acceleration Across All Platforms - Automatically enables Metal and CoreML on Apple Silicon, CUDA on NVIDIA GPUs, and Vulkan or HIP on AMD/Intel hardware for faster transcription.
• Audio File Import and Re-transcription - Import existing recordings in MP4, WAV, MP3, M4A, FLAC, OGG, AAC, MKV, WebM, or WMA format and re-transcribe them locally with any installed model.
• Rich Text Summary Editing - Edit AI-generated summaries in a BlockNote-based editor supporting structured formatting, and use templates tailored to standup meetings, sales calls, project syncs, and more.
• Single Self-Contained Binary - Ships as a single Tauri application with no separate server process to manage, no Docker containers, and no external dependencies beyond the app itself.
• Local SQLite Storage with Migrations - All meeting data, transcripts, and summaries are stored in a local SQLite database with automatic schema migrations and a legacy database import path.

Common Use Cases

• Legal teams documenting privileged client discussions - Lawyers record and transcribe client consultations on their own machines, maintaining attorney-client privilege and meeting data residency requirements without relying on cloud transcription vendors.
• Healthcare providers capturing patient consultations - Clinicians use Meetily to generate local transcripts and structured summaries of patient interactions, keeping protected health information off third-party servers to comply with HIPAA.
• Enterprise product teams protecting proprietary strategy - Product managers and executives transcribe roadmap and strategy meetings locally, preventing sensitive competitive information from being stored on SaaS vendor infrastructure.
• Developers and researchers in air-gapped environments - Engineers working on classified or restricted systems run Meetily in fully offline mode using local Ollama or the bundled llama.cpp sidecar when internet access is unavailable or prohibited.
• Freelancers and consultants managing client confidentiality - Independent professionals transcribe and summarize client calls on their own hardware, avoiding per-meeting SaaS subscription costs while maintaining full data ownership.
• IT administrators deploying meeting tools under data sovereignty requirements - System administrators install Meetily on endpoints to give teams meeting intelligence without creating new data flows outside approved infrastructure.

Under The Hood

Architecture Meetily uses a layered desktop architecture with clear separation between a Next.js/React presentation layer and a Rust core managed through Tauri 2’s typed command system. The backend is organized into distinct bounded domains — audio capture, whisper and parakeet transcription engines, summary generation, database management, and per-provider LLM integrations — each in its own module with minimal cross-module dependencies. An additional process boundary exists for local LLM inference: a llama-helper sidecar built on llama-cpp-2 communicates with the Tauri process over a JSON-over-stdin/stdout protocol, isolating model inference for stability and enabling model hot-switching. SQLite via sqlx handles all persistence with versioned migrations, and Tauri plugins manage filesystem access, notifications, OS detection, and OTA updates. Some legacy artifacts from an earlier architecture remain in the codebase, indicating ongoing but incomplete refactoring.

Tech Stack The desktop runtime is Tauri 2 wrapping a Next.js 14 and React 18 frontend written in TypeScript. The Rust backend targets minimum Rust 1.77 and uses tokio for async I/O, sqlx for typed SQLite queries, and anyhow for error propagation. Audio is captured through platform-native APIs — CoreAudio on macOS and WASAPI on Windows — and processed through FFmpeg for mixing, encoding, and format conversion. Transcription runs via whisper-rs bindings to whisper.cpp, and a custom Parakeet ONNX Runtime integration provides a faster alternative on supported hardware. GPU acceleration is feature-flagged at compile time for Metal, CoreML, CUDA, Vulkan, HIP, and OpenBLAS. On the frontend, Radix UI provides accessible primitives, BlockNote enables rich text summary editing, Framer Motion handles animations, TanStack Virtual powers efficient transcript rendering, react-hook-form and Zod manage form state and validation, and Sonner provides toast notifications.

Code Quality The repository has a limited test footprint — three frontend test files covering BlockNote markdown transformation, summary language preferences, and onboarding model selection logic. No Rust unit or integration tests are present in the cloned source. TypeScript code is well-typed with comprehensive Zod schemas for API boundaries and react-hook-form validation. Error handling in Rust uses anyhow for propagation but Tauri commands return generic string errors rather than typed error enums, reducing structured error handling at the IPC boundary. ESLint is configured for the Next.js frontend; no Rust clippy or formatter enforcement is visible in CI. The overall architecture is disciplined but quality tooling coverage is uneven, and the presence of legacy files suggests ongoing cleanup work.

What Makes It Unique Meetily’s most distinctive technical contribution is offering two swappable local transcription engines — whisper.cpp and Parakeet via ONNX Runtime — within a single application, automatically accelerated by whatever GPU backend the build target supports. The llama-helper sidecar decouples llama.cpp inference from the main process via a lightweight JSON protocol, improving stability and allowing users to run fully offline without Ollama installed. The combination of simultaneous dual-channel audio capture with FFmpeg mixing, VAD-based silence trimming, and a rich text editor for structured summary editing represents a notably complete local-first meeting workflow that most alternatives split across multiple tools or services. This entire stack runs as a single binary with no external server, making deployment genuinely zero-configuration.

Self-Hosting

Meetily is released under the MIT License, one of the most permissive open-source licenses available. You can use it commercially, modify it, distribute it, and incorporate it into proprietary products without any copyleft obligations. There are no restrictions on enterprise or commercial self-hosting, no license keys required, and no audit clauses — the license imposes only the requirement to retain the original copyright notice. This makes it genuinely suitable for regulated industries that need to evaluate and approve all third-party software licenses before deployment.

Running Meetily yourself means taking on the full operational burden of a desktop application rather than a server service. The application installs as a standard desktop binary on macOS and Windows, with Linux requiring a build from source. There is no server infrastructure to manage — all data lives in a local SQLite database in the application data directory. You are responsible for keeping the application updated (OTA updates are built in but require user action), managing local storage growth as meeting recordings accumulate, and ensuring that the hardware running Meetily meets the memory and GPU requirements for the transcription models you choose. Larger Whisper or Parakeet models provide better accuracy but require more RAM and disk space; the team recommends at least 8 GB RAM, with 16 GB for larger models.

Meetily Community Edition is deliberately kept free and fully functional for local transcription and summarization. Meetily PRO, available as a separate paid product at meetily.ai/pro, targets users and teams who need higher transcription accuracy, speaker diarization, custom summary templates, advanced export formats (PDF, DOCX, Markdown), automatic meeting detection, and calendar integration. A self-hosted deployment option for teams is also in the PRO roadmap. If you are evaluating Meetily for an organization with more than a handful of users or compliance requirements around audit trails and guaranteed SLAs, the PRO tier or the enterprise offering at meetily.ai/enterprise is worth reviewing, as the Community Edition has no centralized management, no shared team workspace, and no dedicated support channel.