v1.7.5

NousResearch/hermes-agentv1.7.5Feb 14, 2026by FranckyB

AI Summary

Focused on output formats, metadata embedding, and persistent settings management. Also included bug fixes for UI and audio recording.

Key Highlights

  • Support for WAV, FLAC, and MP3 output formats
  • Embedded metadata in audio files using ID3/Vorbis comments
  • Persistent advanced settings saved per tool and engine
  • "Review Before Saving" mode

New Features

  • Multiple output formats (WAV, FLAC, MP3)
  • Embedded metadata in audio files
  • Persistent advanced parameter storage
  • Review Before Saving mode

Full Release Notes

Version 1.7.5 - Bug Fixes & Polish

Persistent Settings Improvements

- Per-Model-Type Storage - Advanced parameters now save separately for each Qwen model type (Base, Speakers, Trained, Design) instead of a shared "qwen" bucket
- Independent Restore - Switching between model types restores the correct saved parameters for each one
- Accordion Visibility Fix - Fixed advanced parameter sliders resetting to defaults when switching engines, by toggling wrapper visibility instead of accordion visibility

Bug Fixes

- Fixed Return Value Mismatches - Corrected early-return paths in Voice Clone and Voice Design that returned fewer values than expected, causing Gradio errors when validation failed
- Voice Changer Recording Error - Gracefully handles Gradio's microphone recording bug (short recordings causing Content-Length errors) with a clear status message instead of a crash

UI Polish

- Shorter Tab Names - Trimmed tool tab labels to prevent overflow when many tools are enabled

February 14, 2026
Version 1.7.0 - Output Formats, Embedded Metadata & Persistent Settings

Output Format Selection

- Multiple Output Formats - Choose between WAV, FLAC, or MP3 (320kbps) for all generated audio in Settings
- Review Before Saving - New "Review Before Saving" mode lets you listen to results before committing; a Save button appears on each tool to save when ready
- Consistent Save Flow - Save button is disabled until audio is generated, and disables again after saving to prevent duplicate saves

Embedded Metadata

- Self-Contained Audio Files - Generation metadata (engine, seed, text, settings, etc.) is now embedded directly inside audio files using industry-standard tags (ID3 for WAV/MP3, Vorbis comments for FLAC)
- No More .txt Companion Files - Eliminates the need for separate metadata text files alongside each output
- Backward Compatible - Output History still reads old .txt metadata files for previously generated audio
- Powered by Mutagen - Pure Python audio tagging, works on Windows, Linux, and macOS

Persistent Advanced Settings

- Settings Survive Restarts - Advanced TTS parameters (temperature, top-k, top-p, repetition penalty, etc.) are now saved per tool and restored automatically
- Per-Engine Storage - Each engine's parameters are stored independently, so switching engines preserves your tuning
- Automatic Save - Parameters save instantly when changed, no manual action needed

Voice Design Improvements

- Auto-Fill Save Name - When using a Prompt Manager-generated JSON instruction, the save dialog automatically uses the label field as the suggested filename

Output History

- Multi-Format Support - Output History now lists WAV, FLAC, and MP3 files