Twinny changelog

Vision model support.

• Better file context for @ mentions, works in editing message and normal editor.
• Fix some other bugs
• Enable virtual scrolling of messages to help with UI lag when gets too long.

Style and providers overhauled.

Added default providers.

Integrated https://github.com/token-js/token.js/tree/main/src

Removed function calling for now as it's useless and there are better things to do.

Twinny now supports basic function calling.

https://www.twinny.dev/blog/function_calling_released

https://ollama.com/blog/tool-support

1. Fixes embeddings and vector search
2. Update to latest lancedb package
3. Fix some stupid embedding related bugs
4. Smoke something illegal in the UK
5. Fly to the moon
6. Put a flag on it
7. Sit back and watch the world burn
8. Contemplate life
9. Die alone for lack of oxygen

Latest release, refactoring etc.

Full screen chat.

Fetch open PRs and get assistance for code reviews using your preferred chat provider.

• @ mentions for all files in workspace for prompt context
• Large code clean up and refactor
• Initial support for full screen mode
• Fixed bugs with embedding and other thins

Edit, delete and regenerate messages.

• Add tiptap and use @ workspace for embedding context per message
• Add options to specify embedding size.

RAG context and workspace embeddings.

This release is broken for mac and windows.

Symmetry - Inference Network

Symmetry is an experimental decentralized computing network that allows users to share and access computational resources. Initially integrated with the twinny VSCode extension, Symmetry has potential to become a powerful tool for developers, researchers, and data scientists.

https://github.com/twinnydotdev/symmetry.

Save previous conversations and retrieve them later.

Adds a new UI to add, edit and delete custom providers rather than having to open the settings menu every time to change what provider is currently being used.

• Add ability to generate commit messages directly to terminal from staged git changes.
• Added new keyboard shortcut for generate commit message CTRL+SHIFT+T to CTRL+SHIFT+G.
• Refactored chat layout remove selected code toggle

• Enable fully configurable API for both chat and fim endpoints.
• Remove defaults causing headache, user can add their own
• Adds support for custom fim template
• Add LiteLLM support

• Support automatic multiline completions
• Enable multiline completions by default
• Keep option to disable it
• More sophisticated methods to determine multiline completions

Updates Ollama chat completions to work from the OpenAI chat specification /v1/chat/completions.

This is a minor update which affects the chat completions API.

If you were previously using /api/generate or /api/chat for Ollama chat completions please change it to /v1/chat/completions or it will no longer work.

Better context Style updates Bug fixes Other features.

• Add new document button to code blocks
• Fix some style issues with code blocks
• New button styling in code blocks
• Added LMStudio support
• Automatically set port and path when selecting provider

• Add and edit custom templates
• Choose default templates for chat window

Major refactor types, event handlers and more.

• Added support hosted llama.cpp servers
• Added configuration options for separate FIM and Chat completion server endpoints as llama.cpp server can only host one model at a time and fim/chat don't work interchangeably with the same model
• Some settings have been re-named but the defaults stay the same
• Remove support for deepseek models as was causing code smell inside the prompt templates (need to improve model support)

Enabled cancellation of model download when starting twinny and an option to re-enable it.

1. Add option to click status bar icon to stop generation and destroy stream
2. Add max tokens for fim and chat to options.

• Major refactoring and maintenance.

Add option to disable neighbouring tabs file context as completions can vary due to the model getting confused.

Fix and create prompt context from related open files.

• Add completion cache

• Improved inline completion handler, fix double end bracket closing.

• Enable inline completions in the middle of text, this was purposely stopped but it does add value and more Copilot like.

Refactor and clean up.

Add TLS/https support for Ollama API.

Add stop generating button for messages.

• Deepseek coder support

• Persist chat history
• Add option for new chat session

• Disable chat and send button when generating.