Gobi Desktop Changelog

1. Gobi Desktop 3.0 Coming Soon
   Gobi Desktop 3.0 is going to be more integrated experience with ease of use for PKM builders. At the same time, it provides advanced settings for power user to customize their AI experience and PKM experience. Plus, it will have community features so that PKM builders can share their knowledge with others. Stay tuned!

Gobi Desktop 2.0 TODO Items

• Combining/refactoring UI/UX for Settings (global settings, orchestrator settings, voice settings)
• Checking for Claude login status ([Issue](https://github.com/anthropics/claude-code/issues/1886 "https://github.com/anthropics/claude-code/issues/1886))
• Multiple parallel orchestrator session
• Settings folder default cleanup
• Bug reporting skill & skill management
• Terminal Rendering issue after clear on Windows
• JSON output streaming in AI4PKM repo - this helps with updating Claude progress in real-time so that experience is less boring
• JSON input streaming in AI4PKM repo - this helps latency because AI4PKM for orchestrator can be up and running all the time streaming input to input stream instead of spawning a process every single message
• ElevenLabs TTS
• Mute TTS
• Killing ai4pkm for saving tokens (Issue)
  • In conversational mode - sometimes, if I say (or type) multiple messages (before ai4pkm respond), some messages get dropped by claude. (will get worse when we start to kill ai4pkm)
• Langauge option for Orchestrator
• Language option for STT
• Chat history

2026-1-3

• AI4PKM library self-updating
• End-to-end streaming mode, AI4PKM, TTS, STT for faster conversation experience

2026-1-2

• Incorporate append-system-prompt-file argument for system prompt instead of plaintext
• Bugfix - microphone mute is in-app mute not system-wide mute
• Bugfix - new chat should cancel all ongoing mode

2026-1-1

• Improve TTS streaming strategy better
• Add Claude CLI path to customize which CLI user can use

2025-12-31

• Elevenlabs TTS, STT streaming
• AI4PKM input/output streaming
• End-to-end streaming

2025-12-30 Part 2

• Windows Open Obsidian bug
• fix-path to inherit system-wide path from bash config

2025-12-30

• Hotkey configuration in App Settings page
• Fixed Claude Powershell Set-ExecutionPolicy issue
• Small indicator in Vault icon does not get updated when enable/disable vault
• Add wake word behavior -> Passive / Conversational / Manual
• Manual end smartturn detection toggle
• Prefer system-wide claude/codex/gemini over bundled, but for node/python use bundled
• Play start sound effect for start recording for all modes

2025-12-29 Part 2

• New chat --> Cancel existing processes
• Unzipping criterial change from version change (major, minor, patch) to only major or minor
• Running ai4pkm in app terminal
• Mic ON/OFF button (mic mute/unmute)
• Cancelling STT should not be done on "voice_start" (new sentence detected in passive/ambient mode) or sending chat

2025-12-29

• Disable killing ai4pkm when canceling operation. There is [issue with resuming session in early termination of previous session](https://github.com/anthropics/claude-code/issues/7004](https://github.com/anthropics/claude-code/issues/7004)
• Temporarily disable "Claude Auth" button
• Add TTS ON/OFF toggle, and Sound ON/OFF toggle
• Langauge select for Orchestrator language (ai4pkm language)
• Language option for STT
• Add version number to Settings page

2025-12-28

• Added FOR AI4PKM message on splash
• Chat session management change
  • Refreshing session in app launch --> Persist session and "New chat" button to reset
  • Chat log (AI/History) is now {date}-{sessionUUID}.md instead of {date}.md per session
  • Chat entry inside the chat log doesn't have sessionUUID anymore, only {User/Gobi}|{timestamp}>
• Add "Claude Auth" button that generates auth token for 1 year expiry. It opens a Claude login website automatically. No need to run claude from terminal after first install. Click this button.
  • It's always there because [there is no way to check if claude is logged in or not.](https://github.com/anthropics/claude-code/issues/1886)
  • There is open issue for [failing to refresh auth token.](https://github.com/anthropics/claude-code/issues/12447#issuecomment-3625344644)
    • But because the button generates token safe for 1 year, should be okay for now
• Removed Orchestrator mode (passive/conversational) and added a button for passive. Now orchestrator has 3 communication mode
  • Passive: listening and transcribing only (STT -> ai4pkm) + optional periodic processing
  • Conversational: smartturn detection and continuous conversation (idle --> STT -> ai4pkm -> TTS --> back to idle)
  • Manual: one turn conversation (STT -> ai4pkm -> TTS)
• Voice files are saved to userData folder with 7 day retention.
• Added markdown format hyperlink resolution in chat message boxes and message bubble.
• Bug fixes for recording button behaviors
  • Robust cancelling behavior
  • "Manual mode" now use VAD before running STT (reduce hallucination)
• Block opening AI4PKM window when it's in splash screen