Newelle

• Workspaces: Divide chats into workspaces, and tie profiles to them. Different workspaces can run in different windows
• Subagents: You can now create subagents "template" to tackle specific tasks. The main assistant can launch that subagent
• New Long Term Memory: Added a new explorable long term memory that manages memory across chats much better
• Usage Tracking: You can now track how much you use Newelle and specific models
• Better search: You can now search for messages in conversations and options in settings
• Drafts: You can now create drafts of messages

Minor improvements:

• Newelle can now host an MCP server with Newelle's tools for third parties to access
• Improved profiles: They now have explicit inheritance systems and you can transfer settings
• Added support to Qwen Image 2.1 and other image generation tools
• Add a universal screenshot tool
• When downloading local models, their size compatibility is estimated in the model downloader
• Improvements to the voice pill, still experimental
• Long Term Memories can now expose mini apps
• Performance improvements when deleting a message
• You can now delete the chat you are in

• Add voice pill: launch Newelle with newelle --voice to show a pill for fast voice interaction (Experimental)
• Fix no-argument tool calls failing when arguments is an empty string
• Fix Anima image model not loading
• Allow adding MCP servers by pasting their JSON
• Add the option to manually add scheduled tasks
• Fix generate_image tool failing on custom OpenAI-compatible servers due to a hardcoded model name
• Fix OpenAIEmbeddingHandler.get_embedding_size() returning the wrong dimension, breaking FAISS index creation
• Add more statistics like caches and time to first token
• Fix building llama.cpp freezing when there is no internet connection
• Add support for screen recording anywhere on Linux, instead of only on GNOME
• Add support for video input / audio input / file input for OpenAI-compatible providers
• Add support for choosing whether to use RAG or CAG for OpenAI-compatible providers

• New Model Library UI and terminal management.
• Improved builds for llama.cpp, whisper.cpp and stablediffusion.cpp.
• Improved download manager and build cancellation.
• Various bug fixes and stability improvements.

Major update with improved agent capabilities, terminal integration, MCP support and usability improvements.

• Improved terminal tool: Newelle can now manage persistent Terminal sessions and interact with TUIs
• Added mode switching: users can create custom modes that quickly change prompts, tools and skills, such as Planning mode
• Added a curated catalog of MCP servers that can be used to connect applications to Newelle
• Added citations, allowing the LLM to state the source of information
• Added the ability to add LLM providers directly from the UI, including OpenAI, Ollama and Anthropic-compatible providers
• Improved text generation animation
• Added Compact Mode for tool calls and the input bar
• Added the ability to download and explore new extensions and skills directly in Newelle
• Added a Skill editor to create and edit Newelle Skills directly in Newelle
• Added a send_message tool for sending messages to interfaces, such as Telegram
• Added new tools: sleep and switch_mode
• Agents can now read images using the read_image tool
• Added the ability to delegate vision chats to a secondary LLM
• Added support for Mistral STT and TTS
• Added API key support for Ollama
• Improved the Anthropic LLM handler
• Added the ability to set the maximum number of tool iterations in a chat
• Fixed bugs with the OpenAI handler
• Added support for the OpenAI /responses endpoint
• Improved the Telegram interface
• Improved the GUI-API interface
• Improved logging support in the OpenAI API interface
• Fixed bugs with LaTeX rendering and improved performance
• Added an option to automatically expand thinking
• Fixed thinking scrolling up while streaming
• Improved mini-window mode
• Added CTRL+G to switch between normal and mini-window modes