Connecting Captions Brings Privacy-First Live Translation to Mac

Overview

A new open-source tool called Connecting Captions is bringing privacy-first, on-device live translation to macOS users. Designed to run entirely locally, the application leverages Apple’s native translation technologies to provide real-time subtitles without sending audio or text data to external servers. This approach addresses growing concerns regarding data sovereignty and surveillance in professional and personal settings.

The software integrates with Apple’s Theater feature, allowing captions to appear over presentations, video calls, or any active window. By keeping speech recognition and translation processes on the Mac itself, the tool ensures that sensitive conversations remain confidential. It supports a wide array of languages, including Arabic, Chinese, French, German, Japanese, Spanish, and many others, making it a versatile solution for bilingual environments.

Technical Architecture and Privacy

The core philosophy behind Connecting Captions is a "privacy-by-design" architecture. By default, all processing occurs locally on the device. The application utilizes FluidVoice for speech recognition and Apple’s built-in translation engine for converting spoken language into text. Neither the audio input nor the resulting transcripts are uploaded to the cloud unless the user explicitly opts to enable a cloud-based speech model.

"Voice never leaves the machine unless someone turns on a cloud speech model. Theater doesn’t call home. No account. No telemetry."

This local-first approach means the application does not require user accounts, nor does it collect telemetry data. For organizations handling proprietary information or regulated communications, this eliminates the risk of data leakage through third-party servers. The software functions offline once the necessary language models are downloaded to the disk, ensuring reliability even in disconnected environments.

The development team emphasizes that this solution is particularly relevant for "theater mode" usage, where captions are overlaid on slideshows or video conferencing feeds. Because the processing is local, users can share their screen during Zoom or Google Meet calls while displaying translations to an audience, without the underlying audio data leaving their machine.

Supported Languages and Use Cases

Connecting Captions currently supports 25 languages, covering major global languages such as English, Spanish, French, German, Italian, Portuguese, Russian, Chinese, Japanese, and Korean, as well as regional languages like Arabic, Hebrew, Hindi, Turkish, and Vietnamese. This extensive list allows for bidirectional translation in diverse settings, from international business meetings to academic conferences.

The application is tailored for three primary use cases:

1. Presentations and Theater: Users can run Theater over their slides during a presentation. When sharing a screen via Zoom or Meet, the entire screen share will include the captions. Alternatively, users can share only the slides window to hide the captions from the audience.
2. Bilingual Rooms: In spaces where multiple languages are spoken, the tool facilitates immediate understanding. Participants can speak in any of the supported languages, while others view the translated captions in real time.
3. Private Conversations: The Mac can be placed in a room to capture and translate ongoing dialogue. This is useful for multilingual households or collaborative workspaces where immediate translation is needed without recording or transmitting voice data.

For IT departments, the tool offers a practical solution for maintaining compliance with data protection regulations such as GDPR. Since no data is transmitted to external services, the risk of non-compliance with cross-border data transfer rules is significantly reduced.

Installation and System Requirements

To use Connecting Captions, users must have a Mac running macOS 15 or later. Additionally, if the user intends to utilize the Theater streaming feature, the machine must be powered by an Apple Silicon chip. This hardware requirement ensures sufficient computational power for real-time speech recognition and translation tasks without relying on cloud resources.

The installation process is straightforward, available through two main methods. First, users can download the notarized .zip file from the project’s GitHub repository. After unzipping, the application is dragged into the Applications folder. Upon first launch, users must grant microphone permissions within the Theater interface and then press the "Listen" button to begin capturing audio.

"Unzip Connecting-Captions.zip and drag Connecting Captions to Applications. Open Theater, allow the microphone, then press Listen."

Alternatively, users with Homebrew installed can add the developer’s tap and install the application as a cask. This method is preferred by many power users and system administrators who manage software deployments via command line. The commands involve tapping the repository, trusting the source, and installing the cask.

Commercial Licensing and Support

While the application is available for free for personal use, the developers behind Connecting Captions offer a structured commercial licensing model for business and enterprise customers. The named license is priced at $60 per Mac per year, with a minimum purchase of five Macs. This licensing model allows IT departments to scale the deployment across teams without negotiating individual quotes.

For organizations requiring guaranteed response times and formal service level agreements, an SLA add-on is available for an additional $90 per Mac per year. Additionally, a one-time $1,500 fee grants access to an MDM (Mobile Device Management) package, streamlining enterprise deployment across large fleets of devices.

The developers have been transparent about their monetization strategy. They explicitly state that transcripts, voiceprints, and training data derived from user interactions are not part of the sale. This commitment reinforces the privacy credentials of the product, distinguishing it from many other AI-powered translation tools that rely on data harvesting for model improvement.

Security vulnerability reports are addressed within a seven-day window for all users. Paid license holders can further enhance their security posture by adding the written SLA, which likely includes more stringent support and update commitments. The project is developed by Chris Swim Lee under the local-host.ai banner, reflecting a broader movement toward independent, privacy-respecting software solutions in the AI era.