Jot Sie
On-device dictation for Windows — for PCs where the Microsoft Store is turned off.
Download Jot Sie 1.4.12 (.msi, 105 MB)New in 1.4.12
- Lighter and faster. Jot uses about half the memory (around 0.8 GB instead of 1.8 GB), and a long dictation no longer leaves it holding 4 GB until you restart it. Recordings no longer pile up in memory, so hour-long dictations and imports stay light.
- Esc saves without pasting. Esc during a dictation stops it and saves it to your history without pasting it anywhere; the pill still has a Copy button. Esc during a rewrite cancels it, including while you’re still speaking the instruction, and leaves your text as it was.
- Choose what Jot records (since 1.4.11). Settings → General → What Jot records: Everything the mic hears (the default: a call on your speakers is recorded too, as on the Mac), Only my voice (other apps are muted while Jot records, then turned back on), or My voice and the computer's sound (a Zoom or Teams call is recorded straight from the computer, even through headphones).
- PFB models: Claude Haiku 4.5 and GPT-6 Luna (since 1.4.10). A model PFB no longer offers switches to Haiku 4.5 by itself.
- Rewrite handles big selections (since 1.4.10). Rewriting pages of text no longer times out after 30 seconds; a reply cut short is never pasted over your text; Esc cancels a rewrite in progress. A rewrite's page shows Rewritten, Original and Instruction as tabs.
- Steadier on more PCs (since 1.4.10). A missing or busy microphone says what to do instead of "Element not found"; a graphics driver that crashes speech recognition moves Jot to the CPU instead of crashing every launch; picking a new language downloads its model right away.
- Tidier everywhere (since 1.4.10). A pasted API key, a long title or a multi-line selection no longer stretches the page; any single key can be a shortcut.
- New AI page (since 1.4.9). "Prompts" in the left pane is now AI: sign in to PFB and pick a model there, switch on AI cleanup (off by default — after each dictation PFB fixes filler, misheard words, punctuation, numbers and lists), and find your Rewrite prompts, default first.
- Better Rewrite (since 1.4.9). The default Rewrite now also turns spoken punctuation, numbers and dictated lists into written form ("two thirty" → 2:30).
- Rewrite never freezes (since 1.4.8). If an app is slow to hand over the selected text, Jot now gives up after a couple of seconds instead of hanging, and pressing the Rewrite shortcut again while one is running no longer pastes the result into the wrong window.
- Every fix teaches Jot (since 1.4.7). Correct a word any way you like (edit the transcript, Add to Vocabulary, pick the word Jot suggested) and the next dictation writes it right, even names Jot keeps mishearing ("cloud" becomes "Claude"). Each fix shows up as a spelling under the word on the Vocabulary page; remove it there to undo, or choose Keep original to pause it. Jot never replaces words behind your back.
- Updates itself (since 1.4.5): new versions download quietly on the company network and install the next time Jot starts.
- Speaker labels and long recordings (since 1.4.4): … → Detect speakers splits a recording by person; recordings over 7 minutes transcribe.
- Have a version older than 1.4.5? Run this .msi once to update. After that, Jot keeps itself up to date.
Install
- Run the downloaded .msi. It may ask for administrator approval.
- If Windows shows a warning, check the publisher is Vineet Sriram, then click More info and Run anyway.
- Open Jot and follow the setup. It downloads the English speech model once (about 630 MB, from GitHub), then works fully offline.
- Allow the microphone when Windows asks.
Good to know
- Press
Ctrl+Shift+Spacein any app, speak, and press it again — your words are typed at the cursor. - Jot updates itself while you're on the company network. This page always has the latest .msi.
- If your company blocks the installer or the model download, share the error message so it can be fixed.