AirgapAI Transcribe: Private Transcripts and Meeting Notes, Even on Offline Computers

AirgapAI Transcribe now supports Microsoft Foundry Local, Microsoft's runtime and model catalog for running AI on the device in Windows. Support is generally available today and optimized for Intel Core Ultra AI PCs. Support for AMD, Qualcomm and other Arm-based Windows PCs is forthcoming.

With Foundry Local, Transcribe converts speech to text using NVIDIA's Nemotron speech recognition model, which comes from the Foundry Local catalog. Foundry Local picks the version of the model built for the device's hardware (NPU, GPU or CPU), so transcription uses the best accelerator in the machine without any manual tuning. Once the model is on the device, everything runs locally. No audio, transcript or telemetry leaves the device, and no internet connection is needed.

Keep sensitive audio under your control. Summaries are generated by a service that accepts connections only from the computer it runs on, and error messages use plain language without file paths. Report a problem never sends transcript text. Downloaded models that contain program files are rejected, and built-in models are verified and can be restored.

Choose the most accurate option or the most multilingual one without leaving the app. Alongside the built-in Nemotron model, download IBM Granite Speech or Whisper Large V3 from inside the app, where your computer supports them. Each model shows its size and license details so you can decide with the facts in front of you. A model you download is switched to automatically, and a model you chose yourself stays selected.

Whisper Large V3 transcribes in the language spoken. Automatic language detection keeps non-English audio in its own language instead of turning it into English. The detected language is saved with the transcript.

Long recordings read as one continuous transcript. They are processed in longer stretches and stitched together without duplicated or misordered text. For the built-in model, words stay properly spaced and silent tails stay free of stray punctuation.

Know who said what. Speaker labels separate distinct voices and place words at speaker changes with the right person. Rename a speaker once and every mention updates, including after a reload and in exports. Identify speakers works on your existing transcripts, and a Processor use setting lets you trade speed against how much of your computer the work uses.

Turn a meeting into notes without sending it anywhere. Summaries are written on your machine and appear as they are generated, waiting their turn while a transcription is running. Where the summary model is not yet installed, Transcribe can download it in place, and you can cancel the download at any time. Summaries are built from what was actually said, empty sections are dropped, and you can click a point in a summary to jump to the exact passage it came from.

Long recordings stay easy to navigate. Transcripts are grouped into paragraphs by speaker, pause or length, playback highlights the current passage, and Find reaches any part of an hours-long recording. Ctrl+F opens Find, which handles match case, whole words and non-Latin scripts with undo and redo, and Replace all asks for confirmation once before changing anything. Copy text, download the original audio, and keep your own notes beside the transcript.

Take the result into the tools you already use. Save a transcript as Word, PDF, text, SRT or VTT, and export summaries in several formats from a single menu. Summary exports to PDF and Word keep clean formatting, plain-text files are free of stray characters, and Word documents carry a clear title.

Drop a recorded meeting straight into Transcribe. MP4, MOV and MKV video files are accepted, along with a wider range of audio formats. Cancel stops a conversion when you press it, and files in formats such as WMA and AIFF stay playable.

Know your recording is working before you rely on it. A level meter confirms your microphone before you begin, and meeting capture offers a compact device menu with a System default microphone entry. On Windows it can record computer audio alongside the microphone to capture the other side of a call. If a microphone drops mid-take, the take is still saved, and reloading during a system capture picks the recording up again.

A long job is not lost when something goes wrong. A job that loses its model or fails on most of its audio is saved as incomplete with its progress kept, and Start retries only the gaps. Additional files queue behind the current one, cancel works even during conversion, and a recording with no speech says so plainly. When Transcribe is upgraded, your existing transcripts carry over, protected by a backup taken before the change.

Find your way around quickly. The interface is organized around fewer screens, Home combines adding files and browsing transcripts, and Home and Record stay in the header even mid-recording. Add a transcript in one step by choosing a file, dropping it anywhere on the window, or recording, and a single status line shows progress or failure with a Try again option. Search titles as you type and full text after a short pause, then narrow results with one Filter menu and one Sort menu, and restore archived items with undo.

Know where you stand with the trial. Status appears as a banner instead of a blocking window, and deleting files does not restart the trial. When a trial ends, new uploads stop but a recording in progress can still be saved. The trial-ended screen matches AirgapAI Chat and includes a Buy Now option, and each product keeps its own trial clock.

Get answers without leaving the app. Help is a full page with topics, a microphone settings shortcut and a shortcut to your recordings folder. Settings is organized into Basic, Advanced and Trial and license tabs, and changes roll back if a save fails. The Windows installer is leaner, with no test or benchmark data included.

Get started on any computer, including ones with no internet connection. First-run setup recommends a speech model that fits your computer and language and shows the download size before you start. Downloads can be resumed or cancelled and stay visible as you move around the app. On air-gapped computers, import a model by dragging in a file or a folder, and models that are already installed are recognized immediately.