Cleanup levels and the local LLM
Light cleanup is the default and uses on-device rules. Medium and High add a local LLM, so the LLM is off by default. The model stays on localhost. Utterances of eight words or fewer skip the LLM.
What do None, Light, Medium, and High cleanup do?
None leaves the transcript raw; vocabulary and shortcuts still apply. Light is the default and applies rules for fillers, capitalization, and punctuation. Medium adds those rules plus local LLM light polish. High adds rules plus a local LLM rewrite. The LLM runs only at Medium and High, so it is off by default.
On by default: self-correction removal, number normalization, per-app tone, async polish, and signed rule-pack updates from whisperjot.com (download only; nothing uploaded). Off by default: clipboard-as-context.
How do I enable a local LLM for cleanup?
On macOS, runtimes are MLX-LM (Apple silicon, default) at http://127.0.0.1:8080 with model mlx-community/Qwen2.5-3B-Instruct-4bit, or Ollama at http://127.0.0.1:11434 with qwen2.5:7b. Everything stays on localhost. Start with mlx_lm.server --model <id> --port 8080 or ollama pull <model>. Then Settings → Cleanup → level Medium or High → runtime → server URL and model → Test local LLM.
Windows and Linux use Ollama or any local server that speaks a Chat Completions-style HTTP API. MLX is macOS-only. Short utterances of eight words or fewer skip the LLM on every platform.
Straight answers.
Is the WhisperJot cleanup LLM on by default?
No. The local LLM runs only at Medium and High, so it stays off while you leave cleanup on Light, the default, or None. Light uses on-device rules for fillers, capitalization, and punctuation. When you do enable an LLM, it must listen on localhost. Utterances of eight words or fewer skip the LLM even at Medium or High.
Does cleanup upload my transcripts?
Rule-based cleanup runs on your machine. Signed rule-pack updates download from whisperjot.com and upload nothing. A local LLM, if you enable one at Medium or High, stays on 127.0.0.1. Clipboard-as-context is off by default. The opt-in Jot Cloud engine is a separate speech engine and is not how cleanup is delivered.
Can I use MLX-LM for cleanup on Windows?
MLX is macOS-only. On Windows and Linux, point cleanup at Ollama or another local HTTP server, then choose Medium or High, set the server URL and model, and run Test local LLM. On a Mac, MLX-LM is the default runtime at http://127.0.0.1:8080. Keep the server on localhost; the app is not configured here as a cloud cleanup client.
Updated 2026-09-01
Keep reading.
Custom vocabulary and spoken shortcuts
Words correct spelling and can include sounds-like variants. Shortcuts expand a spoken trigger into text and support {date}, {time}, and {clipboard}. Both …
Read article →Per-app tone and cleanup
Per-app tone is on by default. WhisperJot can auto-detect Formal email, Casual chat, Code/verbatim, Structured notes, or General, or you can pin …
Read article →Languages and non-English dictation
The language picker offers Auto-detect plus 15 languages. For non-English speech, use Jot Local Pro (multilingual) or opt-in Jot Cloud. Windows and …
Read article →Ready to dictate?
Download WhisperJot for Mac, Windows, or Linux and start with a hotkey.