Manual model install

Hyprcore normally downloads models on demand from models.hyprcore.ai. If your network blocks that domain, or you're setting up a Mac without network access, you can install models manually.

1. Find your app data directory

The directory is:

~/Library/Application Support/ai.hyprcore.desktop/

Or open Hyprcore → Settings → About and copy the "App Data Directory" path. Or press ⌘⇧D to open the debug menu, which shows the path.

2. Create the models folder

mkdir -p ~/Library/Application\ Support/ai.hyprcore.desktop/models

3. Download the model files

The URLs are publicly accessible. Download from any browser or with curl:

Whisper-family models (single `.bin` files)

Whisper and Breeze both use the Whisper GGML engine, so they're single .bin files.

curl -L -o ggml-small.bin              https://models.hyprcore.ai/ggml-small.bin
curl -L -o whisper-medium-q4_1.bin     https://models.hyprcore.ai/whisper-medium-q4_1.bin
curl -L -o ggml-large-v3-turbo.bin     https://models.hyprcore.ai/ggml-large-v3-turbo.bin
curl -L -o ggml-large-v3-q5_0.bin      https://models.hyprcore.ai/ggml-large-v3-q5_0.bin
curl -L -o breeze-asr-q5_k.bin         https://models.hyprcore.ai/breeze-asr-q5_k.bin

Model	Size	Filename
Whisper Small	487 MB	`ggml-small.bin`
Whisper Medium	492 MB	`whisper-medium-q4_1.bin`
Whisper Turbo	1.6 GB	`ggml-large-v3-turbo.bin`
Whisper Large	1.1 GB	`ggml-large-v3-q5_0.bin`
Breeze ASR	1.1 GB	`breeze-asr-q5_k.bin`

GigaAM (single `.onnx` file)

GigaAM ships as one ONNX file—download it straight into the models folder like a .bin.

curl -L -o giga-am-v3.int8.onnx        https://models.hyprcore.ai/giga-am-v3.int8.onnx

Model	Size	Filename
GigaAM v3	225 MB	`giga-am-v3.int8.onnx`

Directory models (compressed archives)

Parakeet, Moonshine, SenseVoice, and Canary ship as .tar.gz archives that extract to a directory.

# Parakeet
curl -L -o parakeet-v2-int8.tar.gz              https://models.hyprcore.ai/parakeet-v2-int8.tar.gz
curl -L -o parakeet-v3-int8.tar.gz              https://models.hyprcore.ai/parakeet-v3-int8.tar.gz
# Moonshine
curl -L -o moonshine-base.tar.gz                https://models.hyprcore.ai/moonshine-base.tar.gz
curl -L -o moonshine-tiny-streaming-en.tar.gz   https://models.hyprcore.ai/moonshine-tiny-streaming-en.tar.gz
curl -L -o moonshine-small-streaming-en.tar.gz  https://models.hyprcore.ai/moonshine-small-streaming-en.tar.gz
curl -L -o moonshine-medium-streaming-en.tar.gz https://models.hyprcore.ai/moonshine-medium-streaming-en.tar.gz
# SenseVoice
curl -L -o sense-voice-int8.tar.gz              https://models.hyprcore.ai/sense-voice-int8.tar.gz
# NVIDIA Canary
curl -L -o canary-180m-flash.tar.gz             https://models.hyprcore.ai/canary-180m-flash.tar.gz
curl -L -o canary-1b-flash.tar.gz               https://models.hyprcore.ai/canary-1b-flash.tar.gz
curl -L -o canary-1b-v2.tar.gz                  https://models.hyprcore.ai/canary-1b-v2.tar.gz

Model	Size	Archive	Directory name
Parakeet V2	473 MB	`parakeet-v2-int8.tar.gz`	`parakeet-tdt-0.6b-v2-int8`
Parakeet V3	478 MB	`parakeet-v3-int8.tar.gz`	`parakeet-tdt-0.6b-v3-int8`
Moonshine Base	58 MB	`moonshine-base.tar.gz`	`moonshine-base`
Moonshine V2 Tiny	31 MB	`moonshine-tiny-streaming-en.tar.gz`	`moonshine-tiny-streaming-en`
Moonshine V2 Small	100 MB	`moonshine-small-streaming-en.tar.gz`	`moonshine-small-streaming-en`
Moonshine V2 Medium	192 MB	`moonshine-medium-streaming-en.tar.gz`	`moonshine-medium-streaming-en`
SenseVoice	160 MB	`sense-voice-int8.tar.gz`	`sense-voice-int8`
Canary 180M Flash	220 MB	`canary-180m-flash.tar.gz`	`canary-180m-flash`
Canary 1B Flash	950 MB	`canary-1b-flash.tar.gz`	`canary-1b-flash`
Canary 1B v2	1.0 GB	`canary-1b-v2.tar.gz`	`canary-1b-v2`

Silero VAD (required)

curl -L -o silero_vad_v4.onnx          https://models.hyprcore.ai/silero_vad_v4.onnx

4. Install

File models (`.bin` and `.onnx`)

Drop the Whisper and Breeze .bin files and the GigaAM .onnx file directly into the models directory. Don't rename them—Hyprcore identifies them by filename.

~/Library/Application Support/ai.hyprcore.desktop/models/
├── ggml-small.bin
├── whisper-medium-q4_1.bin
├── ggml-large-v3-turbo.bin
├── ggml-large-v3-q5_0.bin
├── breeze-asr-q5_k.bin
└── giga-am-v3.int8.onnx

Directory models

Extract the .tar.gz archives and place the resulting directories—not the archives—in models. The directory name must match the Directory name column in the table above exactly. Parakeet is the one to watch: its archive extracts to a differently named folder.

tar -xzf parakeet-v3-int8.tar.gz
mv parakeet-tdt-0.6b-v3-int8 ~/Library/Application\ Support/ai.hyprcore.desktop/models/

tar -xzf canary-1b-v2.tar.gz
mv canary-1b-v2 ~/Library/Application\ Support/ai.hyprcore.desktop/models/

Warning: Directory names must match exactly. Renaming a model directory breaks detection.

Silero VAD

The Silero VAD file goes in a slightly different place if you're building from source (see the desktop README). For installed app users, it's bundled inside the app.

5. Verify

Restart Hyprcore.
Open Settings → Models.
Manually-installed models appear as Downloaded.
Select one and run a test transcription.

If a model doesn't appear, check that:

The filename or directory name matches exactly.
The file isn't corrupted (compare sizes against the table above).
You restarted Hyprcore after copying the files.

Updating

When Hyprcore ships a new version of a model, the URL stays the same but the file changes. To update, re-download and overwrite. Hyprcore detects the change on next launch.

Was this article helpful?