Release v1.3.0

✨ New Features

• RAG support for local files ( .txt, .md, .markdown, .pdf, .docx, .html, .htm)

🛠️ Improvements

• LLM inference refactor. Splits LLM inference into provider-specific modules and centralizes shared logic
• better dual-embedding (local/cloud) support
• thought retrieval to use the latest user prompt
• adjusted embedding defaults/requirements so local embeddings kick in when no OpenAI key is available for non-OpenAI providers
• assistant-message indexing with a simple filter so only durable responses get stored in the thoughts vector DB
• a guard against empty context injection when no thoughts/summary are available to reduce prompt noise
• a 1200 max chars cap for the thoughts block to avoid prompt bloat from long messages

🐛 Bug Fixes

• fixes OpenAI request build always to have the recently updated conversation data
• fix dev server backend CORS issue
• fix chat history message management
• fix dev server settings write

Full Changelog: v1.2.2...v1.3.0

Release v1.2.2

• Adds Google's API support for speech-to-text and text-to-speech.
• Minor fixes and updates

Full Changelog: v1.2.1...v1.2.2

Release v1.2.1

⚒️ [1.2.1] Bug fixes

This update aims to fix video flickering between avatar states after introducing custom avatars.

Added automatic frame capture so the last rendered video frame is cached before every state swap, refreshed once new media data arrives, and exposed via avatarFallbackImage for the UI.

This ensures seamless transitions without requiring manual poster images and guards against non-DOM environments. Bound the cached frame to the avatar background, providing that the circle shows the most recent frame whenever the video momentarily disappears.

Full Changelog: v1.2.0...v1.2.1

Release v1.2.0

⚒️ [1.2.0] Custom tools and custom avatars support

What’s new:

This update brings more user customization to the Alice app.
You can define your own tools. Automate anything, use Alice to communicate with your local services, or add a new skill that you always missed in Alice.
Read more in docs.

Alice can now change her appearance using custom avatars. Generate a new look for Alice with 3 video-states(standby, thinking, speaking). You can even now convert Alice to Bob or any other avatar.
Read more in docs.

What's Changed

• Bump openai from 5.0.2 to 6.0.0 by @dependabot[bot] in #87
• Bump electron-log from 5.4.0 to 5.4.3 by @dependabot[bot] in #86
• Bump axios from 1.10.0 to 1.12.2 by @dependabot[bot] in #85
• Bump vue from 3.5.13 to 3.5.22 by @dependabot[bot] in #84
• Bump daisyui from 5.0.50 to 5.1.26 by @dependabot[bot] in #83
• Bump dotenv from 16.5.0 to 17.2.3 by @dependabot[bot] in #92
• Bump @tailwindcss/vite from 4.1.4 to 4.1.16 by @dependabot[bot] in #90
• Bump dompurify and @types/dompurify by @dependabot[bot] in #88
• Conversation pipeline refactor by @pmbstyle in #93
• User custom tools support by @pmbstyle in #94
• Custom avatars support by @pmbstyle in #95

Full Changelog: v1.1.10...v1.2.0

Release v1.1.10

⚒️ [1.1.10] Improved local embeddings, markdown, and settings

What’s new:

• Better local embeddings with improved performance and accuracy
• Improved chat markdown by shifting from custom implementation to the Marked library
• Settings now open in a separate window to improve UX
• Some settings have received a styling update
• The Memories manager has moved to the Settings window

Full Changelog: v1.1.9...v1.1.10

Release v1.1.9

⚒️ [1.1.9] Bug fixes

What’s new:

• App update notification styling fix for minified window
• Better markdown parser for links and general tags
• Toggle microphone and speaker hot keys fix

Full Changelog: v1.1.8...v1.1.9

Release v1.1.8

⚒️ [1.1.8] Bug fixes, better local Linux support

What’s new:

• Resolve GPU process crashes on hybrid graphics systems (Linux)
• Improve asset handling and Whisper model management
• Screenshot window losing cursor fix
• OpenAI response ID not found fix
• OpenAI tool call with ID not found fix
• Smaller, compressed screenshots for less token usage
• Remove redundant hardware acceleration disable in Electron

New Contributors

• @tcsenpai made their first contribution in #77

Full Changelog: v1.1.7...v1.1.8

Release v1.1.7

⚒️ [1.1.7] Local Alice – bug fixes, wake-up word, new voices & Searxng support

This release focuses on strengthening local functionality and making Alice more usable out of the box.

What’s new:

• Improved local setup (Windows & macOS ARM):
  More robust handling of binaries and dependencies ensures smoother installs and fewer setup issues.

• Expanded local TTS (Piper):
  Added a range of multilingual voices. Users can now preview and select a voice, and combine it with STT’s language options for full local multilingual support.

• Wake-up word (fixed):
  Now stable and configurable directly from settings.

• New optional web search tool (Searxng):
  Works best with a local Searxng instance.

  Don't forget to allow JSON response format in searxng/settings.yml:

  search:
    formats:
      - html
      - json

  ⚠️ Most of the public instances disable JSON output or block bots/CORS, so reliability depends on your chosen server.

• Linux:
  Still experimental — expect further updates soon.

Full Changelog: v1.1.6...v1.1.7

Release v1.1.6

[1.1.6] Local Alice (BETA*)

This release brings major upgrades to Alice’s independence and usability.

Now you can achieve a completely local setup using:

• LLMs: Ollama or LM Studio
• STT: whisper.cpp
• TTS: Piper
• Embeddings: all-MiniLM-L6-v2

Local and API providers can be combined.
For example, you can use OpenAI for LLM inference while keeping STT/TTS/embeddings fully local.

For local models, the web_search tool (via Tavily) is available, allowing your local LLMs to search and browse the web when needed.

All built-in models run through a Go backend, keeping heavy lifting off the main Electron process and reducing lag.

🗣 Local STT

• Built-in whisper.cpp with model selection, available models:
  • whisper-tiny.en (English only, fastest)
  • whisper-base (multilingual)
  • whisper-small (better accuracy)
  • whisper-medium (high accuracy)
  • whisper-large (best accuracy)

🔊 Local TTS

• Built-in PiperTTS is available for English speech synthesis.
• Available Voices:
  • en_US-amy-medium (default)
  • en_US-hfc_female-medium
  • en_US-kristin-medium

📃Local Embeddings

• Built-in all-MiniLM-L6-v2 that generates a 384-dimensional embedding space.
• OpenAI embeddings(1536 dimensions) and local embeddings designed to work together without data loss

💬 Wake Word Support

With the local STT model, you can now set a wake-up word (like “Alexa”).

• Alice will always listen but only process speech when the wake word is spoken.
• Default mode is auto language detection, but you can also select a specific language in settings.

⚙️ Improvements

• Refactored onboarding wizard and settings for smoother setup and better usability.
• General code readability and stability improvements.

*Local functionality is experimental and can not work as expected, please use v.1.1.5 for stable version

Full Changelog: v1.1.5...v1.1.6

Release v1.1.5

⚒️ [1.1.5] Bug fixes and quality of life updates

• FunctionCaller restructuring
  Separates the logic of functions in dedicated utils/tools files for better maintainability and code readability.

• Run websockets server only if browser_context tool is active
  If we are not using a Chrome extension with the browser context tool, we don't want to start a websockets server.

• Do not TTS links
  Fixes the issue when Alice pronounces links from the response. Now it will be ignored.

• Better chat scrolling
  Fixes the issue when a user cannot scroll to the top when the stream is generating a response.

• Better error display
  Highlights a chat error better with dedicated styling.

• Disable web search and image gen for GPT-5 with minimal reasoning
  Addresses GPT-5 family models' restrictions on using web search and image generation tools with minimal reasoning. Low and higher reasoning will have those tools.

Full Changelog: v1.1.4...v1.1.5