1. GitHub Tool - GitHub API integration, now part of the unified web search tool. 2. Share Sheet - Share files into Llamazing from other iOS apps with the system share sheet. 3. Live Tool Calling - Tools now run mid-reply in a single turn, with no extra decision step — considerably faster, without giving up much quality. 4. Widgets - Redesigned to match the app’s visual language. 5. Download Tool - A new tool that downloads files of many kinds from trusted sources. 6. Tool Cards - Clearer presentation of tools in chat. 7. Text Field - Design improvements to the message composer. 8. Llamigo - New living mascot. You can turn it off in Settings. 9. Voice Chat - Now continues in the background. 10. Copy from Chat - Copy chat text directly, with Markdown adapted for iOS paste.

1. Reasoning Effort – Control how deeply the model thinks with effort levels from Minimal to Max on supported models. 2. OpenAI-Compatible Endpoint – Fixes and improvements for streaming, tool calling, and model compatibility. 3. Web Search – Faster searches with smarter query optimization and source filtering. Model Selector – Fixed a crash when opening the model menu while the chat text field was focused. 4. Scrolling – Smoother, more predictable scroll behavior in long conversations. 5. Chat History – Fixed Markdown syntax showing in chat preview snippets. 6. Sidebar – Improved performance when opening the left sidebar with large chat histories. 7. Projects – Refreshed visual design for the projects section in chat history. 8. Tool Calling – Improved performance and reliability for native tool execution. 9. Token Bubble – Expanded mode now shows a detailed context breakdown with a refreshed design. 10. Chat Navigation – Fixed a slight scroll jump when switching between message sets in a chat. 11. OpenAI-Compatible 12. Endpoint — Video Input – Send video attachments to models and providers that support native video input. 13. Pages Tool – Generate rich-text documents in RTF format, compatible with Apple Pages and Microsoft Word. 14. Develop Tool – Improvements to multi-file project generation, editing, and workspace navigation. Connection 15. Status – More reliable server connection monitoring and status updates. 16. Chat Typography – Default chat font updated to a serif typeface for improved readability. 17. Generation Stats – Fixed high latency between response completion and generation statistics appearing.

Spreadsheet Tool – A full suite of spreadsheet tools for creation, editing, analysis, filtering, and export. Chart Tool – Generate charts and data visualizations directly in chat from spreadsheets or raw data. ComfyUI Image Count – Specify how many images to generate; Llamazing automatically adjusts the workflow batch size. ComfyUI Resolution – Request specific output dimensions; Llamazing applies the resolution directly in the workflow. LFM2.5-2.6B – Added support for Liquid AI's new LFM2.5-2.6B on-device model. Agentic Navigation – A new tool that lets the agent interact with websites by navigating, clicking, filling forms, and completing tasks. ComfyUI 3D GLB Viewer – Textured GLB 3D models from ComfyUI workflows now render in the interactive viewer. File Creation Tool – The AI can create and export files in many formats, including executables, scripts, images, documents, and more.

1. OpenAI-Compatible Endpoint – Connect to any server with an OpenAI-compatible API. 2. Local Inference – Run popular edge-optimized models directly on your device, fully offline. 3. Model Avatars – Model family icons now appear across the model selector and Live Activity. 4. Live Activity (iOS 26+) – Keep LLM generation running in the background indefinitely with the new continued-processing task. 5. Saved Prompts – Fixed prompts that could disappear and fail to open the contextual menu. 6. Presentation Tool – New slideshow agent with templates, layouts, image search, and Keynote/PPTX export. 7. Agentic Mode – When enabled, the assistant can chain multiple tool calls to complete complex tasks. 8. 3D Workflows – Image-to-3D pipelines with an interactive model viewer right in chat.

1. Code Blocks – Now fully personalized with custom colors and styling options. 2. SVG Rendering – SVG blocks now have significantly improved presentation with live rendering. 3. Web Search – Web search and fetch can now fully read Reddit content and YouTube video information. 4. Web Images – Images are now extracted from websites during web search and fetch, and the LLM can display them in its answers when relevant. 5. YouTube Search – YouTube search results now appear as a sortable item in chat history. 6. Tool Calling – Tool calling is now significantly faster and more reliable. Each tool call displays a brief explanation of the action being performed. 7. Tool Call Cards – Tool call cards are now expandable, revealing detailed information about each process. 8. Chat History – Redesigned with a more modern visual concept. 9. Markdown Render – Improved for a better visual and usability experience. 10. Text Field – Redesigned for an improved user experience. 11. Model List – The model selection popover now displays the quantization level of each model. 12. Tool Call Phase – Decision-making phase is now faster. 13. Model Library – Direct access to the model library from the model selector popover. 14. Model Catalog – Improved search functionality and matching logic. 15. Supertonic 3 – Added compatibility with Supertonic 3. 16. LM Studio Thinking – Added thinking mode toggle for LM Studio 0.4.19+. 17. TTS Voice Test – Added voice preview for TTS models that support multiple voices. 18. General UI – Various interface improvements and performance upgrades for a smoother experience.

1. Native Tool Calling – Complete replacement for the Decision Assistant. The AI now directly calls tools (web search, image generation, MCP) in a single request, eliminating the extra LLM call. This results in significantly faster responses, reduced resource consumption, and a more fluid experience. 2. Clarification Tool – New intelligent tool that allows the AI to ask for clarification when faced with ambiguous requests. The AI presents multiple-choice options, and you simply tap your preferred choice to continue the conversation seamlessly. 3. Mermaid Diagrams – Full support for rendering flowcharts, sequence diagrams, class diagrams, and other Mermaid visualizations directly in chat. Diagrams are rendered with automatic error correction and can be exported as PNG images. 4. Custom STT Models – Added support for offline speech recognition using Sherpa-ONNX. Download and use Whisper, Moonshine, SenseVoice, and Paraformer models for high-quality transcription without internet dependency. 5. Multilingual Kokoro TTS – Compatibility with Kokoro-ONNX multilingual models for natural text-to-speech in multiple languages with improved prosody and voice quality. 6. Voice Chat Performance – Significant optimizations in voice chat processing, including memory management improvements and more efficient model loading/unloading to prevent crashes during extended sessions. 7. Bug Fixes – Multiple stability improvements and interface refinements across the entire application.

Embedding – When changing the embedding model, projects are now automatically reprocessed using the new model. Interface – Embedding models no longer appear in the main chat model selector. RAG – Significant improvements in scoring and relevance of attached documents. Interface – Multiple improvements to file attachment and RAG/embedding experience. LM Studio – Fixed a bug that prevented LM Studio embedding models from being displayed. Generation – Fixed a bug that prevented users from canceling generation in some cases. Rendering – Fixed a bug in rendering certain types of HTML (special thanks to Eliad Weizman).

LM Studio — Full LM Studio compatibility, including exclusive features. Model Library — New Hugging Face catalog: search models and download directly to Ollama or LM Studio. Agentic Search — A more powerful web search mode with iterative research and critical source analysis. Markdown — Improved rendering for SVG and HTML blocks for a more faithful chat display. Localization – The app is now available in Spanish, Italian, and French.

1. Fixed a bug that prevented the new image UI from appearing in the text field on some iOS versions. 2. Improved auto-scroll logic so it only activates when the scroll button is tapped. 3. Improved intelligence of the Decision Assistant and the plan orchestrator. 4. Minor UI tweaks and refinements. 5. New quality selector on the voice model download page. 6. Faster extraction of voice model files. 7. Improved Markdown rendering.

1. Minor bug fixes. 2. Initial support with a new UI that lets you paint a mask over attached images. Both the image and the mask are automatically applied to compatible workflows. 3. Improvements and fixes for better reliability.

1. Generation – Fixed an issue that could truncate tokens in responses generated at higher speeds. 2. Model picker – Redesigned the chat model selection menu, now showing details for each available model. 3. Push notifications – You can now receive Firebase push notifications.

1. MCP (beta) – Add MCP servers in-app and let the Decision Assistant call their tools as part of the flow. 2. Architect (beta)– Multi-step planning: the Decision Assistant can create a plan from a single request and execute steps in a queue until the goal is reached. 3. Multimodal context – The Decision Assistant now understands the full chat context, including media, enabling workflows like editing previously shared images via ComfyUI. 4. Parameter bar – Options such as custom prompts are now also applied to the Decision Assistant. 5. Live Activities – Real-time ComfyUI generation progress on the Lock Screen and Dynamic Island (when available). 6. Widgets – New Llamazing widgets for quick actions and fast access to recent/pinned chats. 7. ComfyUI – Improved failure placeholder to surface generation errors more clearly. 8. Automation – The decision model now also coordinates auxiliary tasks, including chat naming and web search handling. 9. Settings – Improved visual hierarchy and better organization across preferences. 10. Media – Fixed a bug that could cause generated results to appear more than once. 11. ComfyUI – Share workflows: export and import Llamazing workflow packages (.llamazing-workflow) with all settings, making it easy to move across devices without reconfiguring from scratch. 12. Performance – Massive optimization to chat rendering: windowed loading + pagination for smoother scrolling and lower memory usage.

1. Voice chat – New hands‑free, live voice conversation mode with the AI (continuous listening + spoken replies). 2. Performance – Major performance optimization for older devices. 3. Stability – Bug fixes and reliability improvements. 4. UI – Visual refinements and polish across the app. 5. ComfyUI – Prompt enhancement now uses attached image context to generate a more specific and intelligent final prompt (recommended: use a vision-capable model in the Decision Assistant). 6. Math – Improved formula rendering logic for better consistency and readability. 7. Documents – Better use of attached document context for more accurate answers. 8. Shortcuts – App Shortcuts integration: open Llamazing straight into Voice Chat to start a quick conversation.

Models – You can now install and remove models on the server directly from the app. ComfyUI – Improved model unload logic before starting a generation, reducing GPU usage and avoiding conflicts. History – New ‘Unload models’ button in the chat history tab. Memory – Optimized system: removed thinking segments from older messages in the context to make processing more efficient.  Keyboard – Fixed a bug that could prevent the keyboard from opening on the first tap in the text field. Performance – General performance optimizations across the app

Decision assistant now explains the decision taken and the next steps, making the flow clearer and more predictable. Improved assertiveness and action selection of decision assistant. New smart auto‑scroll, reactivated when you tap the 'scroll to bottom' button. General – Minor fixes and improvements.

- Fixed an issue that could show cancellation messages when the app went to background even without an actual cancel. - Improved the mechanism that checks whether a model supports thinking. - Fixed a rare slowdown while typing. - Fixed a desync between the connection state shown in the main mini card and in the popover. - Interface improvements for playback and recording. - Fixed an error that could fail to detect the device's native language. Thanks to Eliad Weizman for identifying and reporting the issue. - Fixed an issue where titles could be generated between the decision assistant step and the final answer, slowing the flow. - Added a connection verification step when returning from background.

Projects – Fixed a bug that could prevent correct retrieval of context across conversations within a project.  UI – Fixed an issue where the 'scroll to bottom' button would not take the user to the end of the page during generation.  ComfyUI – Improvement to the background generation notification system.

- Projects – Unified ecosystem: organize your chats into folders and share context and documents across all chats in the project. - Token counter – New option to display the chat’s total token count. - Visual – Liquid Glass effect applied to app elements on devices running iOS 26+. - Layout – Increased horizontal usable space for AI responses. - ComfyUI – The resolution of generated media is now shown in the parameters after generation. - SVG – Improved rendering in conversations. - Thinking – Window now starts open by default showing the live thinking process. - Background – Improved processing and recovery; recover ComfyUI content even after closing the app. - Performance – General app and Markdown rendering improvements. - UI – Fixed an issue that could prevent the “scroll to bottom” button from appearing. - Code – Added a button to copy all code blocks from a message. - PDF – Fixed an issue where bold text inside tables was not formatted correctly when exporting a chat. - Decision Assistant – Fixed an issue where the original prompt language could be changed for web searches. - ComfyUI – Generation placeholder now shows per‑workflow ETA based on previous runs. - Markdown – Added support to render SVG code blocks. - Performance – Substantial improvement when there are many chats and heavy files in history. - PDF export – Better formatting, preserving Markdown styling. - Models – Added “Refresh models” button to the model selector. - Model Selector – Grouping and segmentation by model families. - Audio – Fixed an issue where audio message content was not passed to generate question suggestions. - UX – Improved snap behavior when the user sends new questions, keeping the view available while the AI develops the answer. - PDF Forms – New document filling: send a fillable PDF and ask the AI to complete the fields. Inside a project, the AI can use shared context to fill personal information or previously discussed content.

- Changelog Window – Users can now open the list of changes for each version directly in the app settings. - Drag & Drop – Support for image files, PDFs, and .JSON workflows for ComfyUI. - Automatic Release of the Decision Assistant Model – Added an option to unload the model before generating content with ComfyUI, reducing GPU load and speeding up generation. - ComfyUI Connection Timeout – Adjusted to a more suitable value, preventing premature disconnections. - TTS in ComfyUI – It is now possible to download MP3 files generated by workflows. - Audio Message – Users can send voice recordings, making interaction with the AI faster and more natural. - WebSearch Sources Card – Fixed a bug that prevented proper display. - Markdown – Resolved an error that interfered with text formatting. - Decision Assistant Improvement – Increased accuracy in web searches. - Embedding & RAG System – Now shows the excerpts used for each answer, improving transparency.

- Now, the copy‑text button automatically deletes the thought content. - Fixes a bug where models without thoughts could be identified as thought models, causing request errors. - Minor UI improvements

- Video support added to ComfyUI – users can now play and manipulate video files directly within the interface. - IPv6 crash fixed – resolved the issue that caused the app to freeze when connecting over IPv6 in certain scenarios. - Command substitution bug fixed – the “/” auto‑complete in the text field now works correctly on all devices. - Significant startup performance boost – optimized server connection handling, cutting initialization time by ~40 %. - General UI and performance improvements – fine‑tuned interface responsiveness and fixed minor bugs.

- Fixed bugs that caused old conversations to carry over context into new chats. - Added the ability to edit custom prompts. - Added the ability to edit a ComfyUI workflow. - Improved translation accuracy and coverage. - Implemented ComfyUI command cards that appear when typing “/” in the text field. - Enhanced overall stability and performance.