Zackriya-Solutions/meeting-minutes: Local AI Meeting Transcription in Rust

Cloud-based meeting transcription tools have become ubiquitous, but they come with a hidden cost: your data. For developers, security-conscious enterprises, and anyone handling sensitive conversations, sending audio to third-party servers creates compliance nightmares and genuine privacy risks. The IBM 2024 data breach report puts the average cost at $4.4 million per incident. GDPR fines have reached €5.88 billion by 2025. California alone saw 400+ unlawful recording cases this year.

Zackriya-Solutions/meeting-minutes offers a fundamentally different approach. This Rust-built, open-source meeting assistant captures, transcribes, and summarizes meetings entirely on your local machine. No cloud upload. No data leaving your infrastructure. With 25,079 GitHub stars and an MIT license, it has gained significant traction among developers who refuse to trade privacy for convenience.

What is Zackriya-Solutions/meeting-minutes?

Zackriya-Solutions/meeting-minutes—branded as Meetily—is a self-contained desktop application built with Tauri, combining a Rust backend with a Next.js^{↗ Bright Coding Blog} frontend. The project is maintained by Zackriya Solutions, a team explicitly described as "expert AI engineers passionate about data sovereignty and open source solutions."

The tool sits at the intersection of several active technical domains: on-device machine learning, speech-to-text inference, and local LLM integration. Its 25,079 stars and 2,519 forks (as of the last commit on 2026-06-05) place it among the more prominent Rust-based AI tools on GitHub. The primary language is Rust, which matters for performance-critical audio processing and cross-platform deployment.

Meetily's positioning is deliberate: it targets "professionals and enterprises who need to maintain complete control over their sensitive information." This includes defense consultants, legal professionals, healthcare providers, and enterprise executives—anyone for whom a data breach or compliance violation carries severe consequences.

The project distinguishes itself from cloud alternatives through four documented pillars: privacy-first local processing, cost-effectiveness via open-source models, offline functionality, and self-hosting flexibility. It also offers a commercial Meetily PRO tier with enhanced accuracy and team features, though the Community Edition remains free and open-source under MIT license.

Key Features

Local-First Processing. Every operation—audio capture, transcription, and summarization—runs on your hardware. The README explicitly states: "No data ever leaves your computer." This is not a partial-local hybrid; it's complete data sovereignty.

Real-Time Transcription with Multiple Engine Support. Meetily supports both Whisper (OpenAI's open-source speech recognition model) and NVIDIA Parakeet for live transcription. The Parakeet integration is specifically noted as "4x faster," suggesting significant optimization work for latency-sensitive use cases. Both models run locally without API calls.

Speaker Diarization. The system can distinguish between different speakers in a meeting. This feature is flagged for enhancement in the PRO version with "automatic speaker separation" planned for mid-June release.

AI-Powered Summarization with Provider Flexibility. Summaries are generated through your choice of backend: Ollama (recommended for fully local operation), Claude, Groq, OpenRouter, or any OpenAI-compatible endpoint. This flexibility lets organizations use existing AI infrastructure rather than being locked into a single vendor.

Professional Audio Capture. The application captures both microphone and system audio simultaneously, with "intelligent ducking and clipping prevention"—features typically found in broadcast software, not meeting tools.

Hardware Acceleration. GPU support is platform-optimized: Apple Silicon via Metal and CoreML on macOS; NVIDIA CUDA and AMD/Intel Vulkan on Windows and Linux. The build system auto-detects available hardware, requiring no manual configuration.

Cross-Platform Desktop Application. Native packages for macOS (Apple Silicon .dmg) and Windows (x64-setup.exe), with Linux supported via source builds.

Use Cases

Compliance-Regulated Industries. Healthcare organizations subject to HIPAA, financial services under SEC/FCA recording rules, and legal practices with attorney-client privilege requirements can transcribe meetings without creating third-party data exposure. The local processing model eliminates the need for data processing agreements with external vendors.

Defense and Government Contracting. Organizations handling classified or controlled unclassified information (CUI) often operate in air-gapped or restricted network environments. Meetily's offline functionality and lack of cloud dependency make it viable where internet-connected tools are prohibited.

Intellectual Property Development. Engineering teams discussing patentable inventions, pharmaceutical researchers in early-stage drug discovery, and startup founders in fundraising conversations all benefit from eliminating cloud audio storage. The risk of competitive intelligence leakage through transcription provider breaches or training data harvesting is removed.

Remote Work in High-Risk Jurisdictions. Professionals working from countries with data localization requirements or surveillance concerns can use Meetily without routing audio through foreign servers. The custom OpenAI endpoint support also allows organizations to keep summary generation within jurisdictional boundaries.

Cost-Conscious Teams at Scale. Organizations with frequent, long-duration meetings can avoid per-minute transcription API costs. The open-source model approach shifts costs to hardware—a one-time or amortized expense rather than recurring SaaS fees.

Installation & Setup

Windows

1. Download the latest x64-setup.exe from Releases
2. Run the installer

macOS

1. Download meetily_0.4.0_aarch64.dmg from Releases
2. Open the downloaded .dmg file
3. Drag Meetily to your Applications folder
4. Open Meetily from Applications folder

Note that macOS requires Apple Silicon (ARM64); the specific filename meetily_0.4.0_aarch64.dmg indicates no Intel build is currently distributed.

Linux

Linux requires building from source. The project provides two documentation paths:

• Building on Linux
• General Build Instructions

Quick start:

# Clone the repository
git clone https://github.com/Zackriya-Solutions/meeting-minutes

# Navigate to frontend directory
cd meeting-minutes/frontend

# Install Node.js dependencies
pnpm install

# Build with GPU acceleration support
./build-gpu.sh

The build-gpu.sh script handles hardware detection and enables CUDA, Vulkan, or Metal backends automatically. You need Rust and Node.js installed before running these commands. The Tauri-based architecture means the Rust toolchain compiles the audio processing core while the Next.js frontend handles the UI.

Real Code Examples

The README contains limited explicit code snippets, reflecting its orientation toward end-users rather than library consumers. The primary code example is the Linux build sequence shown above.

Example 1: Linux Build with GPU Acceleration

git clone https://github.com/Zackriya-Solutions/meeting-minutes
cd meeting-minutes/frontend
pnpm install
./build-gpu.sh

This sequence demonstrates the project's hybrid Rust/Node architecture. The git clone pulls the full monorepo. The cd meeting-minutes/frontend navigation reveals the Tauri application structure: Rust backend code lives at the repository root, while the Next.js frontend is isolated in its own directory. pnpm install resolves JavaScript^{↗ Bright Coding Blog} dependencies for the UI layer. Finally, ./build-gpu.sh triggers the Tauri build process with hardware acceleration enabled—this compiles the Rust audio capture and inference code against available GPU libraries.

Example 2: Custom OpenAI Endpoint Configuration

While not presented as a code block in the README, the documented custom endpoint feature implies configuration through the UI or a settings file. The screenshot at docs/custom.png shows interface elements for specifying:

• Base URL for your endpoint
• API key
• Model selection

This enables integration with self-hosted inference servers (e.g., vLLM, TGI, or local OpenAI-compatible proxies), allowing organizations to route summary generation through internal infrastructure rather than external APIs.

The README does not currently provide API-level code examples for programmatic usage. Developers looking to embed Meetily's capabilities into larger workflows should examine the Rust source directly or contribute documentation for library-style consumption.

Advanced Usage & Best Practices

Model Selection Strategy. The choice between Whisper and Parakeet involves trade-offs not fully quantified in the README. Whisper has broader language support and community validation; Parakeet offers the documented "4x faster" performance. For English-only meetings where latency matters, Parakeet is likely preferable. For multilingual or less-common languages, Whisper's extensive training data may yield better accuracy.

GPU Memory Planning. Running transcription models locally requires sufficient VRAM. The README does not specify minimum requirements, but developers should verify their hardware against Whisper and Parakeet model sizes before deployment. The build system's auto-detection helps, but cannot create memory that doesn't exist.

Summary Backend Isolation. Using Ollama for summaries maintains complete local operation. Switching to Claude, Groq, or OpenRouter reintroduces cloud dependency for the summarization phase—acceptable for some use cases, but worth documenting in your privacy assessment. The custom endpoint option offers a middle ground for organizations with private inference infrastructure.

Audio Source Configuration. The "intelligent ducking" feature suggests the audio pipeline is non-trivial. For best results in mixed environments (e.g., conference rooms with both in-person and remote participants), test microphone versus system audio balance before critical meetings.

Version Pinning for Reproducibility. The specific macOS filename (meetily_0.4.0_aarch64.dmg) suggests pre-release versioning. Organizations deploying at scale should track release tags and validate checksums, as the rapid development pace (last commit 2026-06-05) may introduce breaking changes.

Comparison with Alternatives

Otter.ai represents the mainstream cloud approach—superior ease-of-use for non-technical users, but complete data exposure. Generic Whisper Web UI implementations offer local transcription but lack the integrated summarization, audio capture engineering, and polished desktop experience Meetily provides. Meetily's specific value is the packaging: it combines transcription, summarization, and audio capture in a single installable application with GPU acceleration, rather than requiring users to assemble components.

FAQ

Does Meetily work completely offline? Yes. All processing runs locally. No network connection is required after initial download and model setup.

What hardware do I need for GPU acceleration? macOS: Apple Silicon (Metal/CoreML). Windows/Linux: NVIDIA (CUDA) or AMD/Intel (Vulkan). The build script auto-detects available hardware.

Is the Community Edition really free forever? Per the README: "Meetily Community Edition remains free & open source forever." MIT license confirms this.

Can I use my own LLM for summaries? Yes. Ollama is recommended for local operation. You can also configure any OpenAI-compatible endpoint.

What's the difference between Community and PRO? PRO offers enhanced accuracy, custom summary templates, advanced exports (PDF/DOCX/Markdown^{↗ Smart Converter}), auto-meeting detection, and team deployment options. Speaker diarization improvements are planned for mid-June.

Is there an Intel macOS build? The current release is meetily_0.4.0_aarch64.dmg (Apple Silicon only). Intel Macs would need to build from source.

How do I contribute or report issues? Open an issue or submit a pull request following the project structure. See CONTRIBUTING.md.

Conclusion

Zackriya-Solutions/meeting-minutes delivers on a specific, valuable promise: meeting intelligence without data compromise. For developers and organizations where privacy is non-negotiable—whether due to regulatory requirements, competitive sensitivity, or operational security—the local-first architecture eliminates an entire category of risk.

The tool is best suited for: security-conscious enterprises, compliance-regulated industries, remote teams in sensitive jurisdictions, and technically capable users who can manage local GPU inference. It is less ideal for users prioritizing zero-configuration setup over data control, or those needing immediate mobile support (currently macOS and Windows desktop only).

With 25,079 stars and active development, the project has demonstrated genuine community traction. The Rust foundation and Tauri packaging suggest maintainable, performant engineering rather than prototype-quality code.

Explore the repository, download a release for your platform, or build from source: https://github.com/Zackriya-Solutions/meeting-minutes

For organizations evaluating broader AI infrastructure decisions, see our analysis of [INTERNAL_LINK: on-device ML deployment patterns] for complementary tooling strategies.