---
title: "Release notes"
summary: "What changed in each Dictate release."
---

# Release notes

Each stable release has a page here: `docs/releases/<version>.md`, where
the version is `YYYY.M.N` (a correction is `YYYY.M.N-C`). The release cut
writes it from the release's changelog file. The site shows the highest
stable version as the latest release.

Every release, betas included, is also on
[GitHub Releases](https://github.com/arcforgelabs/dictate/releases), and its
complete Markdown changelog is in the
[changelog index](https://github.com/arcforgelabs/dictate/blob/master/CHANGELOG.md),
with the audited
[contribution records](https://github.com/arcforgelabs/dictate/tree/master/CHANGELOG/records).

## Releases

- [v2026.10.2](2026.10.2.md) (latest) - Dictating no longer wipes the clipboard.
- [v2026.10.1](2026.10.1.md) - Transcription no longer reports to Microsoft.
- [v2026.9.27](2026.9.27.md) - `npx @arcforgelabs/dictate install` installs again, and per-user Linux installs can update themselves from the app. npx ran the package's update script for every command, so `install` and the app's `update --user` both stopped with the installer's usage text.
- [v2026.9.26](2026.9.26.md) - Dictate starts without a microphone.
- [v2026.9.25-4](2026.9.25-4.md) - About shows the version the package manager installed.
- [v2026.9.25-3](2026.9.25-3.md) - Right Ctrl dictation works on X11 without extra libraries.
- [v2026.9.25-2](2026.9.25-2.md) - The Linux dock and app menu show the Dictate icon instead of a generic microphone.
- [v2026.9.25-1](2026.9.25-1.md) - The packaged engine includes the X11 hotkey library, so a normal desktop session can dictate without being in the `input` group.
- [v2026.9.25](2026.9.25.md) - The dictation list and **Copy last dictation** follow when the words were spoken.
- [v2026.9.20](2026.9.20.md) - `distil-large-v3.5` is selectable on the faster-whisper lane (English only, about one WER point better than `turbo` at the same speed).
- [v2026.9.19](2026.9.19.md) - Dictate is now local-only.
- [v2026.7.4](2026.7.4.md) - NVIDIA Parakeet-TDT English speech-to-text backend (ONNX via `onnx-asr`, no NeMo/torch).
- [v2026.7.2-2](2026.7.2-2.md) - Local (on-device) dictation now streams through overlapping, silence-aligned, prompt-threaded chunks that are merged into the transcript, instead of joining independent 2-second windows.
- [v2026.7.2-1](2026.7.2-1.md) - Local (on-device) dictation now streams through overlapping, silence-aligned, prompt-threaded chunks that are merged into the transcript, instead of joining independent 2-second windows.
- [v2026.7.2](2026.7.2.md) - Local (on-device) dictation now streams through overlapping, silence-aligned, prompt-threaded chunks that are merged into the transcript, instead of joining independent 2-second windows.
- [v2026.6.23](2026.6.23.md) - Made Dictate private by default: an unset provider resolves to the on-device `faster-whisper` backend.
- [v2026.6.20](2026.6.20.md) - Added an App update view that distinguishes the Python engine from the desktop shell, reports install kind, shows stale shell state, and exposes the correct update action for source and packaged installs.
- [v2026.6.18](2026.6.18.md) - Added chunked streaming dictation for local STT backends so longer recordings are transcribed in bounded windows instead of retaining one large capture until push-to-talk release.
- [v2026.6.6](2026.6.6.md) - Added Windows desktop packaging support for Dictate, including Tauri MSI/NSIS bundle targets and a Windows build script that stages the Python engine before bundling.
- [v2026.6.5](2026.6.5.md) - What changed in Dictate 2026.6.5.
- [v2026.6.4](2026.6.4.md) - What changed in Dictate 2026.6.4.
- [v2026.6.3](2026.6.3.md) - What changed in Dictate 2026.6.3.
- [v2026.6.2](2026.6.2.md) - What changed in Dictate 2026.6.2.
- [v2026.5.18-1](2026.5.18-1.md) - What changed in Dictate 2026.5.18-1.
- [v2026.5.18](2026.5.18.md) - What changed in Dictate 2026.5.18.