Release v2.1.21CodeSplash AI

CodeSplash AI v2.1.21

CodeSplash AI 2.1.21 introduces the first shipped phase of in-app voice dictation, letting you speak into key CodeSplash input surfaces instead of typing everything by hand. This release also adds a new Voice settings area, dictation model setup flows, and local on-device speech recognition designed to keep audio processing on your machine.

August 1, 2026

Full releaseAll updates

Highlights

01

Phase 1 voice dictation is now available inside CodeSplash AI

02

Speak into major AI input surfaces, including composers, terminal-related flows, and focused in-app text fields

03

New Voice settings page with microphone and polish model controls

04

Local on-device transcription powered by an offline speech runtime

05

Packaged desktop builds now better validate the native speech runtime for more reliable dictation support

Changelog

What changed in v2.1.21

Features

  • Added phase 1 of in-app voice dictation so you can speak into CodeSplash instead of typing in supported input surfaces.
  • Added a new Voice settings page for dictation configuration, including microphone selection and polish model options.
  • Added dictation model setup flows and UI to help users get the local speech models ready for use.
  • Added dictation support across key app targets, including AI composer experiences, terminal-related input flows, and other focused in-app text fields.

Improvements

  • Moved heavy local speech recognition work off the main process to improve app stability during dictation sessions.
  • Added searchable Voice settings entries so dictation controls are easier to find from Settings search.
  • Expanded the desktop app’s native packaging support for the new speech runtime across macOS, Windows, and Linux release builds.

Fixes

  • Improved packaged build reliability for dictation by validating that the required native speech runtime is present before shipping desktop builds.
  • Reduced the risk of voice features failing at first use in packaged apps by handling platform-specific speech runtime packaging more carefully.