On-Device Transcription With Local AI
Run speech-to-text and AI entirely on your hardware. The LiteScribe desktop Local Mode uses on-device models for transcription and notes on Windows, macOS, and Linux.
On-device transcription runs speech recognition on your own computer rather than in the cloud. The LiteScribe desktop app does this in Local Mode, using an on-device transcription engine plus an on-device AI model for summaries and notes, accelerated on your GPU or run on your CPU. Your audio never leaves the machine and no outside provider sees your transcript. Available on Windows, macOS, and Linux.
On-device means on your hardware
On-device transcription is speech-to-text that executes on the processor in front of you, not on a remote server. LiteScribe ships this as the desktop Local Mode. The transcription engine and the diarization model that separates speakers both run locally, and an on-device language model handles the AI work afterwards. The result is that a recording becomes a labeled transcript and a structured set of notes without any of it being uploaded.
This is distinct from the cloud path that the web and mobile apps use, where audio is streamed or uploaded to a transcription service. On-device processing is the reason Local Mode can keep your content private and work without a connection. It is a desktop capability because on-device inference needs real hardware access that a browser tab and a phone background process do not reliably provide.
Hardware-aware model selection
On-device inference is only useful if it runs well on your machine, so LiteScribe detects your hardware and recommends a model tier that fits. It accelerates on the GPU when one is available, using CUDA on NVIDIA, Vulkan on AMD and Intel, and Metal on Apple Silicon, and it falls back to the CPU when no compatible GPU is present. You can choose a model size from small to extra large to balance speed against quality for your machine.
For the AI step, the same approach applies. The on-device language model that writes your summaries and action items is picked to fit your RAM and VRAM, so the experience stays responsive. Bigger models produce better notes on capable hardware, and a modest laptop still gets usable on-device results on a smaller tier.
The boundary, stated honestly
On-device transcription runs the entire content pipeline on your hardware, and that is the claim we stand behind: your audio is never uploaded, and no third party ever receives your transcript. The models are a one-time download, fetched and verified the first time you enable Local Mode. After that the desktop app reconnects to LiteScribe periodically to validate your plan and remaining minutes, and it pauses new offline work if it has been out of contact for around two days. That account traffic does not include your audio or your transcript.
So on-device is about where your content is processed, not about the app being permanently disconnected. The transcription and the AI happen on your machine; the subscription check happens over the network on a light cadence. Separately, the finished transcript syncs to your own LiteScribe account by default so you can read it on your other devices, and you can switch that off per recording or for everything new in Settings, Privacy.
Why use LiteScribe Local Mode
Runs on your CPU or GPU
Speech recognition and AI execute on your own hardware, with GPU acceleration where available and a CPU fallback.
Local diarization
Speaker separation runs on-device too, so multi-person recordings come back with each voice labeled, all locally.
On-device AI notes
A local language model writes summaries and action items on your machine, so the whole pipeline is on-device.
Hardware-tiered models
The app picks a model size that fits your RAM and VRAM, from small to extra large, so it stays responsive.
On-Device Transcription With Local AI
- 1
Install LiteScribe desktop
On-device transcription is the desktop Local Mode on Pro and Business. Download the app for Windows, macOS, or Linux.
- 2
Download on-device models
Enable Local Mode and download the transcription and AI models. The app picks a tier that fits your hardware. This download is the only networked step.
- 3
Transcribe locally
Record or import a file and the app transcribes and diarizes on your CPU or GPU. No audio is uploaded while Local Mode is on.
- 4
Generate notes on-device
The local language model writes a summary and action items on your machine. The transcript syncs to your account for cross-device access unless you switch that off in Settings, Privacy.
What is private, and what is honest
Privacy on this page means the desktop Local Mode, where your audio is transcribed on your own CPU or GPU and an on-device AI model writes your notes, so your recording is never uploaded and no outside provider ever sees your transcript. The web app, the mobile app, and the cloud path are different: they use a cloud transcription service and cloud AI, so they are not private in this sense. Local Mode is a desktop feature on the Pro and Business plans, available for Windows, macOS, and Linux after a one-time model download.
We will not call the app air-gapped, because it is not. Your audio stays on your device, but the desktop app reconnects to LiteScribe periodically to validate your plan and remaining minutes, and it pauses new offline work if it cannot check in for about two days. That account check carries none of your content. One more thing worth saying plainly: the finished transcript syncs to your own LiteScribe account by default, so you can read it on your other devices. Turn off "Sync new local transcripts to cloud" in Settings, Privacy, or lock an individual transcript, and the text stays here with the audio.
Frequently Asked Questions
More privacy and offline guides
Keep your recordings on your machine
Download the LiteScribe desktop app and turn on Local Mode to transcribe on-device. 120 free minutes every month, no credit card required.