Model Management

List and download speech and embedded LLM models used by cuttledoc

The model commands manage local speech models and embedded GGUF models. Cloud models and Ollama models use their provider's own infrastructure instead.

List models

cuttledoc models list

The command shows the available local speech and embedded LLM model IDs and marks models that are already downloaded.

Speech models

CLI ID	Model	Purpose
`parakeet`	Parakeet TDT 0.6B v3	Fast local transcription, 25 languages
`whisper`	Whisper large-v3-turbo	Broad local coverage, 99 languages

Download one model:

cuttledoc models download parakeet
cuttledoc models download whisper

Download both speech models:

cuttledoc models download all
# Equivalent alias:
cuttledoc models download asr

all and asr download speech models only; they do not download an LLM.

Local speech models require macOS on Apple Silicon at runtime. See Backends before downloading them for a deployment target.

Embedded GGUF models

These models run through node-llama-cpp and can be downloaded by the cuttledoc CLI:

CLI ID	Model
`gemma3n:e4b`	Gemma 3n E4B
`gemma3n:e2b`	Gemma 3n E2B
`mistral-nemo:12b`	Mistral Nemo 12B
`phi4-mini`	Phi-4 Mini

cuttledoc models download gemma3n:e4b

The CLI uses gemma3n:e4b for correction by default. Downloaded GGUF files are stored in a per-user cache; see LLM Enhancement for platform-specific paths and the CUTTLEDOC_LLM_MODELS_DIR override.

Ollama models

Ollama owns its model store, so use the Ollama CLI rather than cuttledoc models download:

Ollama model tag	Model
`phi4:14b`	Phi-4 14B
`mistral-nemo`	Mistral Nemo 12B
`gemma3n:e4b`	Gemma 3n E4B
`gemma3n:e2b`	Gemma 3n E2B

ollama pull phi4:14b
ollama list

Provider IDs are exact: Ollama uses mistral-nemo, while the separately packaged embedded GGUF variant uses mistral-nemo:12b.

Then select the model during transcription:

cuttledoc podcast.mp3 --llm-model phi4:14b

Cloud models

OpenAI speech and LLM models are hosted and require no local download. Configure OPENAI_API_KEY, then select the speech model with --model or an enhancement model through the @cuttledoc/llm API.

OPENAI_API_KEY=sk-... cuttledoc call.wav \
  --backend openai \
  --model gpt-4o-mini-transcribe

See LLM Enhancement for provider detection, correction models, and cache details.

Verify or retry a download

Run cuttledoc models list after a download to verify its status. If a download was interrupted, run the same download command again. For FFmpeg installation failures or model-loading errors, see Troubleshooting.