GuidesMModels & BYOKRunning refinement on-device
Advanced7 min readPro

Running refinement on-device

Rewrite your dictation with a local model, so even the text-cleanup step never leaves your machine.

Speech-to-text is only half the pipeline — the second half rewrites the transcript according to your workspace. That step can also run locally, which is what makes a fully offline setup possible.

SettingsAI ModelsOn-Device
On-device refinement models require Pro or a perpetual licence — unlike speech, where the Small model is free.

What is offered

Nine curated models, all quantised GGUF builds. A representative selection:

ModelSizeSuits
Phi-4 Mini 3.8B2.5 GBRecommended starting point
Llama 3.2 3B2.0 GBModest machines
Qwen 3.5 4B3.1 GBA step up in quality
Llama 3.1 8B4.9 GB16 GB RAM or more
Phi-4 14B9.7 GB16 GB RAM or more
Gemma 4 26B17.0 GB32 GB RAM or more
Start with Phi-4 Mini. Refinement is a constrained task — cleaning up a transcript against a handful of instructions — and a small model does it well. The jump to a 14B model costs a lot of memory for a modest gain.

Sizing it

The model has to fit alongside your speech model and everything else you are running. As a rough guide, allow the file size plus a couple of gigabytes for context. The context window defaults to 16,384 tokens and is adjustable on the same page — lowering it reduces memory use.

The runtime offloads to GPU where it can and falls back to CPU automatically. Local refinement adds latency to every dictation, so watch the total against the sub-second feel you are used to.

When it is not ready

If local mode is selected but no model has finished loading, InkSpoke does not fail — it injects the raw transcription instead. This is deliberate, but it does mean that "my tone settings are being ignored" can really mean "the local model was not loaded".

Refinement failures are soft across the board: if the model errors, you get your raw words rather than nothing. Reliable, but easy to miss — if output suddenly looks unpolished, check the model status before rewriting your workspace.