Step 1
Open the official releases page
Use the whisper.cpp releases page on GitHub. Open the release you want to install and expand the Assets section.
Tutorial
How to set up local Whisper for Therminal
This guide explains how to get the whisper.cpp binary from GitHub Releases and a ggml .bin model from Hugging Face, then connect them to Therminal's local voice feature.
1. What you need
Therminal uses two separate files: the whisper-cli.exe program and a local ggml-*.bin model.
The GitHub Releases page is used to download the executable binary. The Hugging Face page is used to download the transcription model.
2. Download whisper-cli.exe
Step 2
Download the Windows archive
Choose the Windows asset for x64/x86_64 CPUs. The exact name can change between releases: the practical check is that, after extraction, the folder contains whisper-cli.exe.
Step 3
Extract it to a stable folder
For example C:\Tools\whisper or a similar folder. Avoid leaving the file in Downloads or inside an unextracted zip archive.
3. Download the .bin model
On the Hugging Face page, you will find several ggml-*.bin files. For Therminal, you only need to select one of these files as the local model.
If you use CPU mode and want a low-friction start, begin with ggml-tiny.bin or ggml-base.bin.
about 75 MiB
ggml-tiny.bin
The fastest choice for CPU use and initial tests.
about 142 MiB
ggml-base.bin
A good balance between speed and quality.
about 466 MiB
ggml-small.bin
More accurate, but heavier.
about 1.5 GiB
ggml-large-v3-turbo.bin
Much heavier; useful only if you know you want to push quality higher.
4. Configure Therminal
1. Open Therminal.
2. Press Ctrl + / to open the shortcuts and voice modal.
3. In the binary field, select whisper-cli.exe.
4. In the model field, select the .bin file downloaded from Hugging Face.
5. Set the language to en, it, or auto.
6. Save the voice configuration.
7. In the workspace, hold Shift + Alt + Z, speak, and release to transcribe.
5. If something does not work
If local voice does not start, first check that the paths point to real files and not to folders or compressed zip archives.
If transcription is too slow, switch from a heavy model to ggml-tiny.bin or ggml-base.bin.
If GitHub asset names change in future releases, the fixed requirement remains this: Therminal needs whisper-cli.exe and a ggml .bin model.
6. Reference links
The official GitHub releases page is the reference for current whisper.cpp release assets and executable downloads.
The official Hugging Face page for ggerganov/whisper.cpp is the reference for model names and approximate sizes for ggml-*.bin files.