Release notes

Local Hoist v0.0.36

May 24, 2026

Adds speech-to-text replacements, smart dictation correction suggestions, broader Speech to Text configuration, and improved realtime and AI voice behavior.

New

Added speech-to-text replacement rules so commonly misheard words, names, phrases, and technical terms can be corrected automatically before dictated text reaches the composer.
Added smart dictation correction suggestions: when you edit dictated text, Local Hoist can suggest reusable replacement rules for you to accept and apply to future transcriptions.
Added a dedicated Speech to Text settings panel for choosing transcription models and language behavior, tuning cleanup and long-recording stitching, managing replacement rules, and setting recording limits.

Improved

AI voice can now read short fenced code snippets aloud while still summarizing longer code blocks as visible on screen; text and markdown fenced blocks are always read aloud.
Realtime voice sessions can now end cleanly when the assistant invokes its end-session action, and the realtime configuration modal is simpler and easier to scan.
Speech-to-text now defaults to automatic language detection and includes a second Groq Whisper model option for cases where the non-turbo model may work better for accents or non-English speech.
Quick Preferences can now be controlled from the main preferences area while keeping cost information visible.
Collapsed messages now avoid duplicate single-line content and make better use of available message width.