Release history,build notes, no spin.

✨New Features

• AGENTS.md. Added support for AGENTS.md.
• Local Skills. Added partial support for local skills. The limitation, for now, is that skills can only be read and used from the project's working directory and not the global agents skills.
• Cognitive Interrupt. Added visual cues in the chat when a model gets a Cognitive Interrupt.
• Deleted File Feedback. If you delete a file from disk and the file is open in the editor, this will now give clear visual feedback that the file is no longer available. You can still save it because it's still in the editor cache.

✨Enhancements

• Symbols Indexer. Optimized the Symbols Indexer. It's now up to 20-30x faster than before.
• Text Strings. Updated a few text strings with better and clearer wording.
• Symbols Index. Made further optimizations to the Symbols Index. It now includes the number of lines of each file.
• Console Log. Cleaned up the output console log.
• TODO list. Tweaked the look of the TODO list.
• Dependencies. Updated frontend and backend dependencies.
• Connection Endpoints. Updated the default API endpoint to coder-api.zblade.dev and added coder-direct.zblade.dev as a failover endpoint for improved reliability.

🐛Bug Fixes

• Markdown. Found and fixed the high CPU load when rendering Markdown (finally).
• Git Panel. Fixed a bug with the Generate button that could result in it being pressed twice.
• File Explorer. Fixed a regression in the File Explorer that would make a folder within a folder appear as a file until a restart.
• Terminal. Fixed a bug in the terminal where pasted text would show up twice, once from the terminal and once from the web GUI.
• Diff Decorations CPU. Fixed a bug in the editor that would consume way too much CPU if the document was new and not yet accepted or rejected. This was due to repainting the diff decorations on the whole file.
• Chat Reasoning Render. Fixed a rendering bug in the chat when the model is reasoning that would result in code blocks showing up before other text.
• Model Selection. Fixed a bug that caused the application to select the active model twice at launch with 2 different models.
• WebKit Idle CPU. Fixed idle CPU render loops in the WebKit-based renderer.

✨New Features

• AGENTS.md. Added support for loading and reading AGENTS.md if needed.
• Cognitive Interrupt. Introducing Cognitive Interrupt, helping models break out of loops and stale sessions.
• ZDM. Introducing Zaguán Dimension Memory. This is an extension to the current system that gives the model a much better map of what's available to it and how to access and use it correctly.
• Gemini 3.5 Flash. Both instruct and thinking modes now work.

✨Enhancements

• Local Skills. Added support for using load_skill to read local skills from Zaguán Blade.
• Task Management. Enabled the Task Management TODO list creation instructions.
• User Facing Messages. Enabled and enhanced the instructions for user facing messages between tool calls.
• Duplicate Chat Requests. Guard duplicate blade chat requests.
• Tool Selection Instructions. Added very clear instructions about when to use write_file and apply_patch. Some models would choose the tool it already had even if it was the wrong tool.

🐛Bug Fixes

• Task Logic. Fixed logic that was too strict and resulted in models being unable to do their tasks properly.
• WebSocket Auth. Secured blade websocket auth. The API key would sometimes show up in the logs in plain text.
• Nex N2 Pro. This model has been disabled because it's been retired from the provider.
• Model Stability. Fixed many bugs across zcoderd and CoreX to make every single supported model work and be very stable. This included fixing reasoning chunks, thought signatures, updating SDKs, and testing over an 8 hours non-stop session.
• Model Edge Cases. Fixed different edge cases for many different models.
• GitCommit. Fixed a bug in the GitCommit pipeline that could sometimes result in a double message being generated.
• GPT-5.5. Fixed GPT-5.5 reasoning message rendering.
• Long Conversation Continuation. Fixed many bugs related to continuation of long conversations. Models would sometimes go into loops without getting anything done.
• General. Plus many, many bug fixes.

✨Enhancements

• Symbol Index. The symbol index now includes related symbols and files in its indexing.
• Fast Context. Enhanced and optimized what Fast Context gets. This reduces the token usage and makes it much more targeted and smarter.
• Markdown. Replaced our in-house Markdown renderer with streamdown in the AI Chat for lower latency, CPU usage and complexity.
• Local AI. Added automatic detection of an Ollama server running on localhost.
• System Prompt. Added a default system prompt that is used when no custom prompt is available on disk.
• Open Project. Added an Open Project option in the File menu and a button on the Editor canvas for quicker project switching.
• File Explorer. Made the working directory the root of the project. This makes it easier to manage.
• Dependencies. Updated many node modules and crates.

🐛Bug Fixes

• Editor. Fixed a bug in the editor that resulted in the cursor moving to the top when saving a file.
• Exit Logic. Fixed a bug in the exit logic that prevented it from sending the signal to zcoderd.
• Git. Fixed a regression in the Git implementation that resulted in not being able to push the very first commit.
• File Explorer. Fixed a possible regression that would sometimes result in directories being displayed as files.

✨Enhancements

• Symbols Indexer. Created a much stronger system for how the model should use the Symbol Index more effectively.
• Gemini Models. Hardened Gemini model logic so it uses the available tools instead of defaulting to command-line tools.
• Kimi K2.7 Code. Introducing Kimi K2.7 Code and it's now generally available.
• Anti AI‑Slop and Frontend skills. Optimized the Anti AI‑Slop and Frontend skills.
• PWA skill. Enabled a new PWA skill.

🐛Bug Fixes

• MiniMax M3. Fixed many bugs related to Fireworks to get minimax-m3 to work properly and wow they're fast.
• MiMo v2.5. Fixed multiple prompt and logic issues so both regular and Pro versions behave more consistently.
• Mercury 2. Fixed the prompt and the logic for this model so that it behaves more coherently and with fewer mistakes.

✨Enhancements

• Model Selector. The selected model in the Model Selector is now placed in the middle for easier access to other models.
• Symbols Indexer. Updated the Symbols indexer visual cue in the status bar so that it gets initialized earlier and shows what it's doing.
• UI/UX Refactor. Refactored a lot of the UI elements based on best UI/UX practices in 2026.
• Chat Visuals. Tweaked a few visuals to make the chat look more coherent, cleaner and better.
• Font Size. Added the ability to set the font size in the editor and in the chat. You can either go in to Settings or use Ctrl-Mouse Scroll to adjust.
• Shutdown Contract. Added a new contract at shutdown so that zcoderd can summarise the conversation properly.

🐛Bug Fixes

• PDF Viewer. Fixed a bug in the PDF viewer that resulted in only the first page being rendered and displayed.

✨Enhancements

• System Prompts. Tightened up the System Prompts, made them smaller and more targeted based on Prompt Engineering practices in 2026.
• Skills. Tightened up all of the skills and made them a lot smaller. Catalog is now 1,692 lines total, down from the original 7,369.
• Conversation Summarisation. Created a proper summarising system that gets run when the client exits or at a set number of conversation turns.
• Claude Fable 5. Enabled the new model from Anthropic, Claude Fable 5 (docs).
• Claude Fable 5. Updated the logic to support Claude Fable 5, set low effort for git commit generation and reasoning summary for chat.

🐛Bug Fixes

• Nex N2 Pro. Fixed a bug that prevented Nex N2 Pro from successfully applying patches.

✨Enhancements

• Fast Context. Added support for Fast Context. This will only be available from v0.8.0 and beyond.
• Telegram. Added support for using Telegram as long as the application is running on your computer.
• Screenshots. You can now add boxes, text, and arrows with different colors when grabbing a screenshot or a region. Read blog post
• Themes. Added 2 new themes: Nebula Dust (dark mode) and Alpine Frost (light mode).
• GitCommitMessage. Added support for GitCommitMessage that bypasses the normal request pipeline and saves tokens and time.
• Editor Gutter. Changed the line number gutter background color to a solid instead of being transparent and reduced the width.
• Search Modal. Modified the look on the Search modal so that it fits better with the themes and overall look and feel.
• CodeMirror Editor. Optimized and refactored the CodeMirror editor and the React code based on best practices in 2026.
• Visuals. Added the relative path to the file in the chat and the git panel.
• Local AI. Normalized the system prompt naming for Local AI models. For example, gemma4.md is now valid for all Gemma4 models.
• Diff View. When the model writes a new file, it now shows up as a diff that you can accept or reject.

🐛Bug Fixes

• Stop Signal. Fixed a bug that prevented the Stop signal from being sent to Zaguán Coder Daemon.
• Editor. Fixed a bug in the Editor that would result in the document scrolling to the top when a model edited the active document.
• Window Resize. Fixed a bug that made it impossible to resize the application by dragging the window borders.
• AI Chat. Fixed a conflict between the Latest pill and the scroll function that resulted in flickering when scrolling with a mouse.
• Editor. Fixed a bug in the editor where a larger diff would not properly show.
• Local AI. Fixed the pipeline so that we send only the Git Commit prompt when generating the commit message.
• Chat Panel. Fixed a regression where the queue message would show up when sending a regular request.

✨Enhancements

• Qwen 3.7 Plus. Added and enabled Qwen 3.7 Plus. This model is now generally available.
• System Prompt Builder. A complete rewrite of the System Prompt Builder. Reduced initial token usage, easier maintenance, more control.
• Zaguán Coder Daemon. Added 2 new skills: strangler-migration and surgical-integrity.
• Zaguán Coder Daemon. Added full support for versioned Full Context.
• Zaguán Coder Daemon. Updated the system prompt and logic for Deepseek V4 Pro/Flash to make it more reliable and to take action quicker.
• Zaguán Coder Daemon. Properly wired up the new Fast Context logic and added instructions in the prompts.
• Zaguán Coder Daemon. Tightened the prompts and logic for several models to make them behave better and be more concise.
• Nex N2 Pro. Enabled Nex N2 Pro. This model is currently under testing and is free at the moment from Novita.
• Zaguán Coder Daemon. Enforcing the use of write_file/apply_patch for certain models.
• GitCommitMessage. Added support for GitCommitMessage that bypasses the normal request pipeline and saves tokens and time.

🐛Bug Fixes

• Stop/Cancel Signal. Fixed bugs that prevented the Stop/Cancel signal from being sent to the working model.
• Zaguán Coder Daemon. Fixed a bug where zcoderd would send the summary message twice.
• Zaguán Coder Daemon. Fixed bugs in the prompts and logic for the Deepseek and MiMo models.
• Zaguán Coder Daemon. Added a fix for sessions that went stale after a server restart.
• Zaguán Coder Daemon. Added Xiaomi MiMo to tool call retry logic.
• Zaguán Coder Daemon. Fixed bugs that prevented Kimi K2.6 from properly displaying its reasoning.
• Zaguán Coder Daemon. Fixed system prompts and logic so that models get to work faster instead of fetching content forever.
• Cohere Command A+. Fixed bugs that prevented Cohere Command A+ from working. The model is now generally available.
• Inception Mercury 2. Fixed bugs that prevented Inception Mercury 2 from working. The model is now generally available.
• Zaguán Coder Daemon. Disabled Mistral Small 4 for now due to it making too many mistakes during testing.
• Zaguán Coder Daemon. Fixed bugs that prevented reasoning from appearing correctly for Qwen3.7-Max.
• Zaguán Coder Daemon. Fixed bugs that prevented reasoning from appearing correctly for GLM-5.1.
• Zaguán Coder Daemon. Fixed bugs that prevented reasoning from appearing correctly for Kimi K2.7.
• Zaguán Coder Daemon. Fixed bugs in the Responses logic that prevented reasoning from appearing correctly.
• Zaguán Coder Daemon. Fixed bugs in a few model implementations that prevented them from doing tool calls properly.
• Zaguán Coder Daemon. Fixed a bug that prevented the MiMo v2.5 models from generating a Git Commit message.
• Model Reliability Audit. Completed a full model reliability audit. Verified all supported models and fixed multiple issues across Blade, zcoderd, and CoreX.

✨Enhancements

• Chat. It's now possible to stop the chat and edit the last message before resending it. This is useful for fixing typos.
• Editor Tabs. Added a right-click context menu for editor tabs.
• AI Chat. The chat input field now has a history that can be navigated with Arrow Up and Arrow Down.
• Git Panel. Made the Git panel buttons look and feel the same.
• Semantic Search. Added support for Markdown in the Semantic Search database.
• YOLO Mode. Added YOLO mode that can be enabled in Settings->Configuration. This lets models run without interrupting you.
• Themes. Tweaked themes for better contrast, colors and readability.

🐛Bug Fixes

• Git Commit. Fixed an issue when generating a git commit where some models would try to use run_command.
• Git Panel. Fixed a bug in the Git panel that resulted in it being impossible to push the very first commit.
• Internal Commands. Fixed internal commands exit output to be sent to the model.
• Zaguán Terminal. Working on making the Zaguán Terminal coherent and not launch a new instance every time. (WIP)
• AI Chat. Fixed a bug that caused typing lag in the chat input.
• Editor. Fixed a bug in the editor that resulted in a diff view disappearing when the edited line was clicked.

✨Enhancements

• Zaguán Coder Daemon. Updated the System Prompts based on how modern LLMs in 2026 work in terms of positive and negative instructions.
• Zaguán Coder Daemon. Added support for Xiaomi MiMo 2.5 Pro.
• Zaguán Coder Daemon. Cleaned up the list of supported models and removed unused/outdated models.
• Zaguán Coder Daemon. Added and enabled Cohere Command A+. Will be testing it to evaluate its capabilities as a potential featured model.
• Zaguán Coder Daemon. Tuned several system prompts to make them more informative and interactive during tasks.
• Zaguán Coder Daemon. Added Grok Build 0.1 and will evaluate it as a potential featured model.
• Zaguán Coder Daemon. Updated the blindspot-radar skill based on deep research into Unknown Unknowns.

🐛Bug Fixes

• Zaguán Coder Daemon. Fixed a bug that would result in the conversation stopping after the first message for some models.
• Zaguán Coder Daemon. Fixed bugs in the Qwen system prompt and logic.

🐛Bug Fixes

• Hotfix. Fixed critical AI Git commit message generation bug.

✨Enhancements

• Zaguán Coder Daemon. Giving better feedback to the models on tool call errors.
• Zaguán Coder Daemon. Updated 2 skills to be more aligned with 2026 best practices.
• Zaguán Coder Daemon. Added a new skill, complexity-radar, based on Codex Complexity Optimizer.
• Zaguán Coder Daemon. Enabled parallel tool calling for DeepSeek v4.
• Dependencies. Updated Tauri to the latest version.
• AI Chat. Optimized the AI Chat based on technical inspiration from T3Code.
• CodeMirror Editor. Optimized the CodeMirror editor to get React out of the way to make it faster, more reliable and better overall.
• PHP Support. Added highlighting for PHP in CodeMirror.
• Node Modules. Updated Node modules and crates for both frontend and backend.
• Local Storage. Fixed save and retrieval in both Zaguán Blade and Zaguán Coder Daemon.
• Progress Card. Refined the Progress card in the chat by removing unnecessary information.
• Rust Cleanup. Refactored and removed old and almost unused Rust code/files.
• Zaguán Blade. Created a new version of the AI Chat Panel.
• Alt-TAB. Alt-TAB now properly displays the Project name and the Filename.
• CSS. Refactored a lot of the CSS for the UI to use proper names instead of hardcoding values.
• Scrolling. Made scrolling a lot smoother across the board.
• Dependencies. Added .pnpm-store to .gitignore and Gix ignored logic.
• Shutdown Modal. Added a shutdown modal to save unsaved changes.
• Updated Translations. English and Spanish translations updated and hardcoded literals replaced with translation strings.

🐛Bug Fixes

• Zaguán Coder Daemon. Fixed a bug where it reconnected to the provider in a longer session.
• Zaguán Coder Daemon. Wired up zlp_request. It was planned, but missing.
• Zaguán Coder Daemon. Fixed loading the History in Zaguán Blade.
• Zaguán Coder Daemon. Fixed a bug that prevented Mistral 3.5 from working past the first response.
• Zaguán Coder Daemon. Fixed a bug that could result in the model continuing to work after a clear end summary message.
• Zaguán Coder Daemon. Fixed connection bug after a long period of inactivity.
• Zaguán Coder Daemon. Fixed an issue that would block proper run_command tool calls using sed or cat.
• File Explorer. Fixed a bug where moving files was not possible using the mouse.
• Terminal. Fixed a bug where it wasn't possible to interact with the Terminal when the model was running commands.
• Local AI. Fixed a bug in the Local AI implementation that would result in Gemma4 dying after the first response.
• Editor. Fixed a bug that resulted in stale content being shown when switching tabs.
• Markdown Editor. Fixed a bug in the Markdown editor that would result in text being garbled.

✨Enhancements

• New Models. Introducing DeepSeek V4 Pro and Flash, plus gpt-5.5. These are currently in testing and may be unstable.
• New Models. Crafted a new system prompt for gpt-5.5 based on documentation from OpenAI.
• Zaguán Coder Daemon. Added a new PRD Skill.
• Zaguán Coder Daemon. Added a save and fetch of memories.
• Dependencies. Updated node modules and crates to their latest versions.

🐛Bug Fixes

• Zaguán Blade. Fixed a bug that would overwrite .gitignore.
• Zaguán Blade. Fixed a bug that resulted in the Diff View would not appear, nor the Accept/Reject question.
• Zaguán Blade. Fixed a bug in the file state that resulted in apply_patch hanging without being able to complete successfully.
• Zaguán Blade. Fixed a bug that would cause CPU spikes at idle.
• Zaguán Blade. Fixed the issue where the application would go black after a while of inactivity.
• Zaguán Blade. Fixed warnings in tsconfig.json and UseChat.tsx.
• Zaguán Blade. Fixed lag in the chat when sending long messages, especially Markdown.
• Zaguán Coder Daemon. Fixed an issue in the Google Gemini system prompt and logic to make it work better.
• Zaguán Coder Daemon. Fixed parallel tool calling. Gives a perceived 10x speed improvement.
• Local AI. Fixed a bug so that it loads the correct System Prompt.

✨Enhancements

• Zaguán Coder Daemon. Added a brand new skill aimed at doing a deep security audit.
• Zaguán Coder Daemon. Fixed how models communicate and how much.
• Zaguán Coder Daemon. Added the full path and file name so that the model knows what it's working with.
• Zaguán Coder Daemon. Added retry logic for all models in case of different errors occurring.
• Diff View. Enhanced the Diff View in Zaguán Blade and made it look more coherent and nice.
• WebSocket Migration. Moved the History and the ZLP lookup from HTTP/SSE to WebSocket in both zcoderd and zblade.
• Zaguán Blade. Optimizations and enhancements to both frontend and backend.
• CodeMirror. Made the scrolling feel smoother.
• Canvas. Added a hook to repaint the canvas, especially on Linux/X11/Openbox.
• Local AI. Zaguán Blade now supports Gemma 4 with Local AI (Ollama and OpenAI-compat).

🐛Bug Fixes

• General. Many bugs fixed.
• Terminal. Fixed bugs that caused the Terminal to not display Tab text or requests from the shell.
• Terminal. Continuing to fix the bug where run_command requests from the model aren't showing up correctly.
• Local AI. Fixed a Local AI leak that was sent to the server.
• File Explorer. Fixed a regression where files and folders weren't displaying correctly.

✨Enhancements

• Semantic Search. Zaguán Blade now analyzes your codebase and builds a deep reference map that the model can query semantically.
• Smarter Context Retrieval. Instead of pulling broad chunks of code, the model can now pinpoint the most relevant symbols, files, and references it actually needs.
• Lower Token Usage. Semantic search helps reduce context size while improving response quality and precision across code understanding tasks.
• Configuration Redesign. Reworked Settings into Configuration for a clearer and more intentional setup experience.
• Fully implemented i18n for Zaguán Blade, with English and Spanish now available.
• Updated to Vite 8.
• Updated CodeMirror modules to their latest versions.
• Updated many outdated Rust crates.
• Replaced a select few commands with built-in commands for better cross-OS compatibility.
• Zaguán Blade. Disabled a lot more debug output throughout the app.
• Zaguán Coder Daemon. Made the symbol_* tools first-class citizens.
• Harness. Continued strengthening the harness across many levels. Testing is still needed.

🐛Bug Fixes

• Tool Calls. Fixed a bug where the final response could arrive above a run_command tool call.
• Blade Terminal. Fixed a bug where extra text would sometimes precede the command from run_command.
• Release Audit. Completed a final release audit and fixed several lingering issues, including removing the old Chat implementation.
• Chat Panel. Fixed Markdown rendering so messages feel more fluid and natural.
• Chat Input. Fixed a bug where the cursor was rendered at double the font size.
• Editor. Fixed the font color of inactive line numbers in the gutter to be brighter and more visible.
• Diff View. Fixed a regression in the editor Diff View.
• Image Attachments. Fixed a bug where attached images would not be correctly sent to the model.

✨Enhancements

• Zaguán Coder Daemon. Introducing GPT-5.4-Mini with Low and Medium reasoning.
• Local AI. It is now possible to try and use Zaguán Blade with Local AI models.
• First-Run Experience. Improved the onboarding flow with Local AI as a no-setup option.

🐛Bug Fixes

• Tool Calls. Fixed the issue where tool calls would always appear at the top of the conversation instead of inline where they belong.
• Setup Flow. Removed the blocking setup dialog that prevented users from exploring the app.
• Messaging. Clarified the messaging around Local vs Cloud usage.
• UX & Copy. Various UX and copy improvements throughout the app.
• Zaguán Coder Daemon. Fixed a bug where we didn't catch the reasoning chunks correctly for the GLM-5 model.
• Zaguán Coder Daemon. Fixed a bug where the final response from MiniMax M2.5 and M2.7 was placed above the tool calls when it should be the last message in the conversation.

✨Enhancements

• Zaguán Coder Daemon. Implemented a new GPT skill inspired by Uncodixfy and refined the other design skill.
• Zaguán Coder Daemon. Fixed streaming issues related to gpt-5.4.
• Zaguán Coder Daemon. Added a tool_search tool based on the Anthropic documentation to help reduce initial token usage.
• AI Chat Panel v2. A complete rewrite of the AI Chat panel. v2 is now a lot more performant and responsive.
• Chat Panel Update. Updated the Chat panel and took a lot of inspiration from T3Code.
• Git Commit Text Area. Made the Git commit text area bigger and dynamic.
• AI Chat Panel. Toned down the font color to make it easier to read.
• AI Chat Panel. Made completed tool calls fade a little to make it easier to separate.
• Editor Breadcrumb. Minor design tweak to the breadcrumb at the top of the editor for improved visual hierarchy.

✨New Features

• Code and Planning Mode. You can now switch between Code and Planning mode. Added a special Planning skill in Zaguán Coder Daemon to go along with this new feature.
• Implement Button. Implemented an "Implement" button when in Planning mode. Clicking this switches to Code mode and begins the work.
• Themes. Added theme support with four built-in themes: the original dark theme Zaguán Dark plus Blade Coder, Blade Cream and Blade Robot.

🎨Design Updates

• Theme Selector Dropdown. Tightened up and redesigned the design of the Theme selector dropdown menu.

🐛Bug Fixes

• AI Chat Conversation Order. Fixed a bug where responses and tool calls would not arrive and display in a natural conversation order.
• Git Push Lock Issue. Fixed a bug in the Git panel where a git push would complain about a git lock.
• Gitoxide Integration. Moved more Git functionality from git to gitoxide.
• Streaming Logic. Fixed a bug in the streaming logic that would sometimes garble the responses from specific models.
• GPT-5.4 Streaming. Fixed the streaming issue when using gpt-5.4 that resulted in garbled responses.
• Cancelled Tool Calls. Fixed an issue where a cancelled run_command tool call would end up below the response from the model.
• Chat UI. Removed noisy elements from the chat and tightened up some of them as well. Much nicer view.
• Git Commit Generate. Fixed a bug in the Git Commit Generate path that would sometimes fail for certain models.
• Git Commit Generator. Fixed the Git Commit generator to also check for/include new/staged files. This was missed before.
• Blade Terminal Rendering. Rewired the logic to use the Blade Terminal again and hopefully fixed the rendering bugs.
• AI Chat v2 Design. Made minor tweaks to the AI chat v2 design.
• Terminal Resize. Fixed the resize logic for the Terminal—previously wonky, now much smoother.
• Keyboard Shortcut. Added a new keyboard shortcut - Ctrl-L - to put the chat input in focus.
• Diff View Styling. Fixed Diff View styling across all 4 themes. The new themes were using the style from the original theme.
• Zaguán Coder Daemon. Fixed the Gemini continuation thinking bug.
• Zaguán Coder Daemon. Fixed bugs in relation to tool_search that would sometimes result in the model failing to apply patches and then overwrite files.
• Debug Output. Disabled a lot of debug output in the console.

✨Enhancements

• Frontend Audit. Completed a full frontend audit: removed dead files, tightened up many functions, removed unused fonts, and made the codebase cleaner and leaner.
• i18n Preparation. Began preparing the application for internationalization (i18n); English is now ready.
• Zaguán Coder Daemon. Enabled parallel tool calling for supported models.
• Zaguán Coder Daemon. Added preliminary support for Mistral Devstral 2.
• Rust Backend. Did an audit of the Rust backend using the Rust Skill. Found and fixed a few bugs, memory optimizations and panic-prone locks.
• Zaguán Coder Daemon. Working on making the Mercury 2 system prompt stronger and better.
• Zaguán Coder Daemon. Replaced Gemini 3 Flash Preview with Gemini 3.1 Flash Lite Preview.
• Zaguán Coder Daemon. Initial testing support for Inception Mercury 2.
• Zaguán Coder Daemon. Enabled OpenAI GPT-5.4 with low and medium reasoning.

✨New Features

• Zaguán Coder Daemon. Cherry-picked the best pieces from Superpowers and ported them to the GUI environment of Zaguán Blade.

🐛Bug Fixes

• Run Command. Fixed a bug in the run_command card where the command would get detached when it shouldn't.
• Editor Colors. Fixed an issue where the editor did not show colors at all due to incompatible CodeMirror versions.
• Editor Scrollbar. Fixed an issue where the editor horizontal scrollbar was laid over the terminal.
• Layout Alignment. Fixed an alignment issue in the space between the editor and the chat.
• Chat Image Paste. Fixed a bug introduced after the audit where pasting images into chat no longer worked.
• Chat Streaming. Fixed a bug in the chat so that regular responses now stream correctly.
• Chat UI. Fixed the chat and removed a lot of noisy cards.
• New Chat Creation. Fixed a bug when creating a new chat where the model received no data at all.
• Zaguán Coder Daemon. Removed artificial limitation on the number of turns models could do.
• Zaguán Coder Daemon. Fixed a bug where the conversation wouldn't end properly, especially after repeated tool calls.

✨Enhancements

• Markdown Editor. Show a proper Diff view and the Accept/Reject pill in the Markdown editor when it's in Edit mode after a model has made changes.
• Message Queue. Added a message queue so that you can send messages while the model works. These will be executed sequentially.
• Zaguán Coder Daemon. Working on ironing out bugs in several models, Gemini 3.1 Pro, MiniMax M2.5, GPT-5.3-Codex.
• Zaguán Coder Daemon. Disabled several variations of models in the `low` and `medium` ranges to clean up the list.
• Zaguán Coder Daemon. Used the Codex guide to make the implementation better.
• UI. Added the Project Name to the header and to the window title.
• Zaguán Blade. Now sending the local time and date to the model. This helps with year hallucinations and other time related issues.
• Zaguán Coder Daemon. Used gemini-cli source code to get the best information on how to handle it in the best possible manner.
• Zaguán Coder Daemon. Tightened up grep_search command with a timeout and excluding certain directories to speed things up.

✨New Features

• Git Push. Pushing to Git now accepts all changes automatically.

🐛Bug Fixes

• Zaguán Coder Daemon. Fixed a bug that resulted in a warning for the Gemini model. Did not affect functionality.
• Git Push. Fixed a bug in the Git Push button where it now correctly changes state from Push to Pushing to Pushed.
• Header Alignment. Fixed an alignment issue in the header where the app name was off-center when no tabs were open.
• Local AI. Fixed an issue where the responses would appear in reverse order.
• Terminal. Fixed the Terminal Copy command Ctrl-Shift-C/Cmd-Shift-C.
• UI Focus. Fixed a possible bug where the UI wouldn't get focus at launch due to the custom window decorations in Tauri.
• CodeMirror. Fixed a bug in the CodeMirror implementation that would show stale diff views.
• Terminal. Fixed a bug where pasting content into the Terminal would paste twice.
• Chat Panel. It's now possible to resize the chat panel again.
• Terminal. Removed the horizontal resize bar.
• Editor Test. Fixed a test case in the editor.

✨Enhancements

• Project Index Tools. Added 2 new tools to retrieve the project_index.md file. This file can be used by the model to get an overview of the project.
• Editor Context. Send filenames of all open tabs in the editor to the AI model.
• Zaguán Coder Daemon. Dug through a few open-source projects to figure out the best way to make gpt-5.3-codex work seamlessly. A huge thanks to these very smart and talented people building awesome tools.
• OpenAI integration. OpenAI GPT-5.3-Codex is now available in Zaguán Blade.
• Zaguán Coder Daemon. Tightened up the logic for Google Gemini 3.1 Pro. TTFT now 4-5x faster.
• Git Panel. Enhanced Git Panel to show a strikethrough for elements that are moved, renamed or deleted.
• Chat. Added visual feedback while waiting for the model to respond in the chat.

🐛Bug Fixes

• GPT Codex. Fixed a bug where, sometimes, GPT Codex model would emit characters from other languages seemingly at random.
• GPT Codex. Fixed a bug where gpt-5.3-codex would stop after a few tool calls.
• Patching. Fixed a bug in apply_patch_to_string that would replace instead of add new content.
• Local AI. Fixed a bug in the Git commit generation for Local AI where it would fail. This was due to the message being routed wrong.
• Welcome screen. Fixed alignment issue on the Welcome screen.
• Paste image. Fixed a bug where pasting images from clipboard into chat didn't work.
• Tab focus. Active tab now always stays in focus while new tabs are opening.
• Git. Fixed a bug where the .zblade directory would be submitted to git.
• Chat. Fixed a bug where images attached to a message were being re-sent on every subsequent turn, causing context bloat.

✨Enhancements

• Zaguán Coder Daemon. Added stronger detection for Google Gemini if it was unable to apply a patch to help it recover and try again instead of hanging.
• Zaguán Coder Daemon. Added a runtime guard to prevent AI models from inappropriately using shell commands (e.g., grep, cat) instead of dedicated tools, automatically guiding them to correct usage.
• Dependencies. Updated modules for frontend and backend.

🐛Bug Fixes

• Research Progress. Fixed an issue where research progress indicators (like "RUNNING_TOOLS") would remain stuck in a running state when a chat session ended or was stopped.
• Run Command. Fixed a bug in Zaguán Blade where a Run Command, when skipped, would still stay running.
• Run Command. Fixed a bug when the Stop button was pressed and a Run Command was running and would stay running.

✨New Features

• Alpha Tag Removed. Removed the "alpha" tag as Zaguán Blade has reached a usable state.
• UI Redesign. Redesigned the whole application. It's now a lot more professional, calm and better structured.
• Settings. Added an About section in Settings.

✨Enhancements

• Performance. Fixed memory leaks, optimized for better performance and cleaned up the code a little.
• Editor Performance. Significantly improved editor responsiveness by splitting context state and actions to eliminate unnecessary re-renders in hot-path components.
• Markdown Performance. Improved typing responsiveness in Markdown files by disabling expensive syntax features and skipping per-keystroke cursor synchronization.
• Google Gemini 3.1 Pro. Trying to make Google Gemini 3.1. Pro work better based on feedback and other OSS projects.
• Google Gemini. Updated our Google Gemini implementation in Zaguán Coder Daemon.
• MiniMax. Updated our MiniMax implementation for Fireworks in Zaguán Coder Daemon.
• Dependencies. Updated gix to v0.51.0.

🐛Bug Fixes

• Diff parsing. Fixed line-number drift in the editor caused by newline metadata.
• Uncommitted changes. Fixed cumulative diff tracking to properly display changes for files edited multiple times.
• Fixed a bug in the global Accept All/Reject All that was wrongly removed.

✨New Features

• Google Gemini 3.1 Pro. Added Google Gemini 3.1 Pro in Zaguán Coder Daemon.
• Chunk counting. Added chunk counting and display for write operations tool calls.
• Qwen3.5 support. Added support for Qwen3.5 models in Zaguán Coder Daemon.
• Claude Sonnet 4.6. Enabled Claude Sonnet 4.6 in Zaguán Coder Daemon. Restart application to access it.

✨Enhancements

• Performance. Optimized streaming performance and conversation context synchronization.
• Frontend. Updated Vite to version 7.3.1.

🐛Bug Fixes

• Zaguán Coder Daemon provider fix. Fixed an issue where Anthropic-specific instructions were being sent to non-Anthropic providers.
• Fixed a bug when using Gemini 3.1 Pro where extra text would appear when using the run_command tool.
• Fixed a bug when run_command started a command without exit that blocked progress.
• Diff parsing. Fixed an issue where Diff parsing was only reading the first parsed patch block.
• Fixed a warning message that was too prominent.
• Fixed escaped characters showing up in the Blade Terminal.
• Fixed terminal crash with dead-key characters. Resolved UTF-8 unsafe slicing in sentinel parsing that caused crashes with multibyte characters (e.g., ~/dead-key sequences). ANSI stripping now preserves UTF-8 bytes correctly.
• Fixed a bug where the global Accept All/Accept All didn't hide if Accept changes on file was triggered.

✨Enhancements

• When selecting a screenshot, added a half-second delay before capturing to allow time for UI elements (like window selection boxes) to disappear.
• Git performance. Replaced most external git calls with gix for faster, more efficient Git operations.
• Git Commit message. Now uses the full git commit message instead of just the first line.
• Local AI refactoring. Refactored Local AI so that it's separate from the main AI flow for easier maintenance.

✨New Features

• Brand new App Icon. Redesigned app icon with a fresh, modern look.
• Git graph visualization. Implemented a Git graph of the 50 latest commits with a details popup for viewing commit information.
• Ollama Cloud support. Added support for Ollama Cloud. Requires an account at ollama.com.

🐛Bug Fixes

• Fixed text disappearing or flickering when sending new messages after loading a conversation from history.
• Fixed scroll not going to bottom when loading an existing conversation from history.
• **CodeMirror "Ranges must be added sorted" crash.** Three extensions could add decorations in unsorted order causing CodeMirror to crash:
  • diffDecorations.ts — Sort was (from, to) but CodeMirror requires (from, startSide, to). Fixed by tracking startSide on each decoration and sorting by (from, startSide, to).
  • rainbowBrackets.ts — Visible ranges expanded to line boundaries could overlap, causing duplicate/unsorted bracket decorations. Fixed by tracking maxProcessed to skip already-processed positions and sorting all brackets before adding.
  • indentGuides.ts — Same overlapping visible ranges risk. Fixed with maxProcessed guard.
  • Chat UI. Fixed the chat and removed a lot of noisy cards.
  • New Chat Creation. Fixed a bug when creating a new chat where the model received no data at all.
  • Zaguán Coder Daemon. Removed artificial limitation on the number of turns models could do.
  • Chat UI. Fixed the chat and removed a lot of noisy cards.
  • New Chat Creation. Fixed a bug when creating a new chat where the model received no data at all.
  • Zaguán Coder Daemon. Removed artificial limitation on the number of turns models could do.
• Chat UI. Fixed the chat and removed a lot of noisy cards.
• New Chat Creation. Fixed a bug when creating a new chat where the model received no data at all.
• Zaguán Coder Daemon. Removed artificial limitation on the number of turns models could do.
• Chat becomes inert after UI reload. After a UI reload (e.g. crash recovery), the frontend's loading state was false even though the backend was still streaming. Added get_chat_status Tauri command that returns true if the backend is streaming. Frontend init() now calls get_chat_status and restores loading = true if the backend is still active. Files: chat.rs, lib.rs, useChat.ts
• Can't scroll to bottom after crash/reload. After reload, the scroll container starts at the top. The isUserAtBottomRef quickly becomes false from the scroll handler, so streaming auto-scroll stops working. Added an initial scroll-to-bottom effect that fires once when messages first load, ensuring the user sees the latest content and isUserAtBottomRef stays true for streaming auto-scroll. Files: ChatPanel.tsx
• Fixed the Blade Terminal echo bug when the model is using the run_command tool call.
• Fixed run_command tool calls in AI Chat appearing above already run (or failed) commands instead of below them.
• Fixed bug in the Ollama and Ollama Cloud implementation that resulted in errors.
• Diff view showing up on wrong lines. Fixed issue where diff view would appear on incorrect lines due to stale data in the editor. Added proper cleanup and state invalidation when editor content changes. Files: DiffView.tsx, EditorPane.tsx
• Model selector sluggishness. Reduced avoidable work in the compact selector by memoizing derived lists and selected model. Removed delayed/smooth scroll on open (which can feel laggy) and switched to requestAnimationFrame + instant nearest scroll. Files: CompactModelSelector.tsx
• History titles missing. Added a resilient fallback title strategy: use conversation.title if present, else use preview, else synthesize from last active timestamp (Conversation X ago). Files: HistoryTab.tsx
• Natural chat autoscroll + user can interrupt with wheel. Reworked autoscroll to use the real scroll container ref. Split behavior: scroll on new message if user is at bottom / just sent; during streaming, only keep following while user remains at bottom. This preserves manual scroll interruption. Files: ChatPanel.tsx
• Settings save feels slow. Changed save flow to close modal immediately, then persist in background. Keeps UI responsive while save + model refresh continue asynchronously. Files: SettingsModal.tsx
• General snappiness + launch flicker. Removed duplicate no-tab editor rendering path that could cause visual churn. Tightened startup flow to show window after first frame and fade out loading overlay immediately (no extra nested delays). Files: Layout.tsx, main.tsx
• Modernized title bar buttons. Updated window controls to compact rounded buttons with tighter spacing and better active feedback. Files: TitleBar.tsx
• Terminal too close to sidebar shadow. Increased left padding in terminal content area. Files: TerminalPane.tsx
• Fixed crash in read_file_range on out-of-bounds slice. Added bounds clamping to prevent panic when requesting lines beyond EOF (e.g., start_line: 350 on a 259-line file). Now returns safe empty output for extreme or inverted ranges instead of crashing.
• Fixed reasoning blocks being dropped during final answer transition. Changed filter logic to preserve reasoning blocks when clearing stale text content. Files: useChat.ts
• Fixed caret sync issues in Command Center textarea. Added shared autosize helper for consistent height updates and implemented explicit Shift+Enter newline insertion with proper caret position restoration. Files: CommandCenter.tsx

📦Dependencies Updated

Comprehensive dependency upgrade across frontend and Rust ecosystem.

Rust Crates - Safe Updates

Rust Crates - Breaking Updates

• tauri (2.9.5 → 2.10.2)
  No code changes required. Unblocked plugin support and JavaScript improvements.
• tree-sitter (0.24 → 0.26)
  Updated node.child(i) to node.child(i as u32) to match new API.
• notify (6.1.1 → 8.2.0)
  No code changes required. API remains compatible.
• reqwest (0.12.28 → 0.13.2)
  Removed rustls-tls feature (now default in new version).
• rusqlite (0.32 → 0.38)
  No code changes required. API remains compatible.
• tokio-tungstenite (0.21 → 0.28)
  WebSocketConfig now mutable default. Message::Text and Message::Ping take Utf8Bytes and Bytes respectively (converted using .into()).

✨New Features

• PDF support. The editor can now load PDFs natively.
• Screenshot support. Capture window or select region and attach to chat.
• Upload image. Upload images directly to chat.
• Paste image. Paste images from clipboard into chat.

✨Enhancements

• Git implementation is now complete. Full Git integration ready for use.
• Added tooltips on the buttons in the sidebar.
• Adjustments in the status bar. Improved status bar layout and information.

🐛Bug Fixes

• Fixed the Blade Terminal input and output where it used to show escape characters.
• Fixed issue where the chat would sometimes show garbled response.
• Hide Search icon because functionality not yet implemented.
• Open tab now shows the Close button.
• After closing a tab, go to the previously open tab.
• Plus many bug fixes.