By ✦ min read

Developer Launches 'Scripta': First Open-Source macOS Meeting Transcriber That Runs 100% Offline

Breaking News — A software engineer has released Scripta, an open-source macOS application that records and transcribes both sides of a meeting in real-time entirely on the device, with zero data sent to the cloud. The app solves a critical privacy and security gap for enterprises that prohibit meeting audio from leaving corporate networks.

"I spend 2–3 hours daily on Teams and Zoom calls and couldn't remember commitments," said Hwang, the developer. "Cloud services like Otter.ai were blocked by our security policy, so I built Scripta to run everything locally—no subscriptions, no data exfiltration."

Background

Most cloud transcription services, such as Otter.ai, Fireflies, and Granola, process audio on remote servers. For companies with strict data residency rules, this is a deal-breaker.

Existing macOS alternatives often fail to separate mic audio from system audio, causing confusion over who said what. Apple's own speech framework, SFSpeechRecognizer, throws an error (kAFAssistantErrorDomain Code=1101) when attempting to run two concurrent recognition tasks—a problem known as the dual-channel dilemma.

Technical Breakthrough

Scripta overcomes this by using two completely different speech engines. The microphone stream is processed via whisper.cpp with Metal GPU acceleration, achieving over 15x real-time speed (5 seconds of audio transcribed in ~0.3 seconds). System audio from remote participants is handled by Apple's on-device SFSpeechRecognizer, which doesn't compete for GPU resources.

"This hybrid approach side-steps the concurrency crash while keeping everything on-device," Hwang explained. System audio capture is made possible by ScreenCaptureKit, introduced in macOS 13, eliminating the need for hacks like virtual audio devices or kernel extensions.

What This Means

Scripta offers a viable alternative for privacy-conscious organizations and individuals. By running entirely on the local Mac, it eliminates the risk of sensitive meeting data being intercepted or stored on third-party servers.

The app also generates AI summaries using a local Ollama large language model, providing actionable insights without any cloud dependency. "Zero cloud requests, zero subscriptions—it's a complete offline solution," Hwang emphasized.

Availability and Future

Scripta is open-source and available on GitHub. The repository includes the base Whisper model (142 MB) and instructions for setup. As of now, the app supports real-time transcription of both local and remote audio, with AI summaries generated after the meeting ends.

For enterprises, this could mean a shift away from paid cloud services toward self-hosted, open-source tools that give full control over data. The dual-engine architecture also sets a precedent for handling multi-channel audio on macOS without relying on Apple's constrained APIs.