Starstruk Logo MarkStarstruk

1.6.0

langflow-ai/langflow1.6.0Dec 15, 2025by Kieirra

AI Summary

This release introduces major LLM connectivity features for Murmure, including support for local models like Ollama, alongside new text formatting rules and a toggle-to-talk recording mode. It also features a backend refactor, UI improvements, and fixes for transcription pipeline issues.

Key Highlights

  • LLM Connect feature allowing Ollama integration with Murmure
  • New Formatting Rules for automatic text transformations
  • Toggle-to-Talk Mode as an alternative to push-to-talk
  • Sound feedback for recording start and stop actions
  • Experimental macOS build (Intel architecture only)

New Features

  • Connect Murmure to a local LLM (Ollama) with presets
  • Automatic text transformations (trailing spaces, find/replace)
  • Toggle-to-Talk recording mode
  • Sound feedback for recording actions
  • Improved UI notifications
  • Dictionary configuration migration
  • Fixed transcription pipeline issues
  • Experimental macOS build

Full Release Notes

### ✨ New Features
- **(Major) LLM Connect** — Connect Murmure to a local LLM (Ollama) to enhance, correct, or reformat transcriptions through post‑processing. Includes presets for translation, medical dictation, development, and automatic reformulation.
- **(Major) Formatting Rules** — Automatic text transformations: add trailing spaces, insert spaces before `?!`, convert text-to-numbers, or apply custom find/replace rules.
- **Toggle-to-Talk Mode** — New alternative to push‑to‑talk: press once to start recording, press again to stop. Thanks to @nicolascrop.
- **Sound Feedback** — Audio cues when starting and stopping a recording.

### ♻️ Refactor
- Improved UI notifications.
- Backend refactor aligned with SRP principles.
- Dictionary configuration migration in preparation for import/export support. Thanks to @icristescu.

### 🐛 Bug Fixes
- Fixed overlay misalignment with fractional display scaling. Thanks to @flomillot (PR #73).
- Fixed a transcription pipeline issue where processing would stop when the transcription was empty, preventing dictionary rules, formatting rules, and LLM post‑processing from being triggered.

### 🧪 Experimental
- Added an **experimental macOS build (Intel architecture only)**.

### 🙌 Thanks
Special thanks to our beta testers: @HachetAmaury, @Grujowmi, and @aparrot.