v0.14.0

v0.14.0 Release Notes

This release adds the Creations page for AI-generated images, image editing/variations features, OOBE wizard for first-time users, mlxcel inference engine support, disk usage tracking, and many UI/UX improvements.

New Features

• Add Creations page for browsing and managing AI-generated images
• Add image editing (inpainting) feature using /v1/images/edits endpoint
• Add image variations feature using /v1/images/variations endpoint
• Add basic image generation UI in chat interface
• Implement OOBE (Out-of-Box Experience) wizard for first-time users
• Add mlxcel inference engine support with format-specific defaults
• Add comprehensive disk usage tracking and storage management
• Add dockable system metrics panel with slide-out side drawer design
• Bundle CLI binary with desktop app and add shell integration settings
• Add mmproj file download support for vision models in Featured Models section
• Add recommended model placeholder cards for empty state on Models page
• Add separate UI tab in Settings page
• Complete Korean localization and improve English text in Agent Settings
• Replace technical ETA label with consumer-friendly time remaining display
• Bundle Ubuntu fonts and improve cross-platform typography

Improvements

• Implement data-driven mapping tables for GPU/CUDA compatibility detection
• Improve UI cosmetics and layout consistency
• UI improvements and title generation fixes
• Upgrade continuum-router from v0.33.0 to v0.34.0

Bug Fixes

• Creations page bug fixes and design improvements
• Use iTXt chunks for PNG metadata to support UTF-8
• Resolve image generation hang with tauri-plugin-http
• Preserve Gemini thought_signature in tool calls
• Resolve conversation title generation issues
• Prevent scroll bounce on macOS sidebar
• Prevent System Metrics panel from overlapping header
• Restore header transparency by allowing content to scroll behind
• Align model dropdown to the left
• Fix model dropdown positioning in StatusBar
• Filter mmproj-only files from model list to prevent standalone entries
• Detect vision capability immediately after model download
• Display error message in assistant chat bubble on API failure
• Display external endpoint connection errors in Continuum router logs UI
• Detect and validate moved model files instead of triggering re-download
• Runtime installation state becomes stale when files are manually deleted
• Recommend CUDA 13 for RTX 50 series GPUs with driver 570+
• Engine download stuck at 100% Verifying on Windows (regression from)
• Detect vision capability for models with mmproj files
• Resolve persistent notification pulse animation after notifications are read
• Add blob: to CSP img-src for image attachment preview
• Remove duplicate labels in Model Config Drawer toggle items
• Prevent image attachments for non-vision models
• Add missing colorPrimaryBgHover token to dark theme files
• Ensure all local models display after HuggingFace download
• Resolve invisible Save Settings button in ModelConfigDrawer footer

CI/CD Improvements

• Convert OUTPUT_DIR to absolute path in source build script
• Use ubuntu-22.04-arm64 runner for Linux ARM64 build
• Include SYCL SPIR-V files (.spv) in engine package for Intel Arc GPU
• Bundle SYCL native SPIR-V libraries for Intel Arc GPU support
• Add CLI binary build step to packaging workflow

Technical Details

• Resolve ESLint warnings across codebase
• Resolve clippy warnings in CLI project
• Resolve ESLint and TypeScript linting errors across codebase
• Technical reports added for helping humans understand decisions

Dependencies

• Update @tauri-apps/plugin-dialog to 2.6.0
• Update Rust dependencies

Breaking Changes

None

Known Issues

None

Update Metadata

This release contains update metadata JSON files for all channels. Do not delete.

v0.13.0

Summary

v0.13.0 introduces the Glass theme inspired by macOS Liquid Glass design, UI zoom controls, multi-part GGUF model validation, Intel GPU detection with SYCL accelerator recommendations, improved download retry/resume capabilities, and numerous bug fixes for Windows stability and dark mode support.

New Features

• Glass theme inspired by macOS Liquid Glass design
• UI zoom in/out functionality with keyboard shortcuts (Cmd/Ctrl +/-)
• Multi-part GGUF model completeness validation
• Intel GPU detection and SYCL accelerator recommendation
• Download retry/resume UI with auto-retry for transient errors
• Warning dialog when changing models directory with existing models
• Persist last selected model in Chat and Agent menus
• Show checking models message on Chat menu entry
• Linux ARM64 CPU target to engine build matrix
• Source build script for custom llama.cpp platforms

Improvements

• Bump llama.cpp to b7717, continuum-router to v0.33.0, and all-smi to 0.17

Bug Fixes

• Header model status flickers on navigation between pages
• z-index conflict between ToolApprovalDialog and MainLayout overlay
• Settings persistence with atomic writes and retry logic for Windows
• Model directory setting persists and reflects in model list (Windows)
• Engine installation completion event emission
• Missing semantic color tokens in dark theme files
• Notification panel icons and dismiss button alignment
• Notification badge overlay graphics corruption
• Console window flash on Windows during GPU detection and engine operations
• Chart axis and labels show correct colors in dark mode
• Window size and position persistence across app restarts
• Frontend fixes for issues and
• Model list not refreshing after deletion in grouped display mode

CI/CD Improvements

None

Technical Details

• Improved atomic file writes for settings persistence on Windows
• Better event emission for engine installation lifecycle
• Enhanced z-index management for overlapping UI components

Dependencies

• llama.cpp: b7717
• continuum-router: v0.33.0
• all-smi: 0.17

Breaking Changes

None

Known Issues

None

What's Changed

• feat: Add Glass theme inspired by macOS Liquid Glass design by @inureyes
• fix: Model list not refreshing after deletion in grouped display mode by @inureyes
• fix: Frontend fixes for issues and by @inureyes
• fix: Persist window size and position across app restarts by @inureyes
• feat: Implement UI zoom in/out functionality with keyboard shortcuts by @inureyes
• feat: Add validation for multi-part GGUF model completeness by @inureyes
• fix: Chart axis and labels show correct colors in dark mode on Statistics page by @inureyes
• fix: Prevent console window flash on Windows during GPU detection by @inureyes
• fix: Prevent notification badge overlay graphics corruption by @inureyes
• fix: Align notification panel icons and dismiss button properly by @inureyes
• fix: Add missing semantic color tokens to dark theme files by @inureyes
• fix: Emit completion event when engine installation finishes by @inureyes
• fix: Ensure model directory setting persists and reflects in model list (Windows) by @inureyes
• feat: Show checking models message on Chat menu entry by @inureyes
• feat: Add warning dialog when changing models directory with existing models by @inureyes
• feat: Persist last selected model in Chat and Agent menus by @inureyes
• fix: Improve settings persistence with atomic writes and retry logic for Windows by @inureyes
• feat: Improve download retry/resume UI and add auto-retry for transient errors by @inureyes
• fix: resolve z-index conflict between ToolApprovalDialog and MainLayout overlay by @inureyes
• feat: Add Intel GPU detection and SYCL accelerator recommendation by @inureyes
• fix: Header model status flickers on navigation between pages by @inureyes

v0.12.1

Release v0.12.1

This patch release adds default model settings configuration and fixes several issues including Vision API compatibility and Windows engine download prompts.

New Features

• Default model settings for chat, image, and agent interfaces

Improvements

• Consolidate icon components and remove inline styles

Bug Fixes

• Windows: Improved binary not found error messaging and add engine download prompt
• Add default width/height to all icon components
• Use TrashIcon instead of undefined DeleteIcon in conversation menu
• Align delete icon with other menu items in conversation context menu
• Remove duplicate DefaultModelsSettings struct definitions
• Support Vision API content array in socket chat message type
• Preserve Vision API content format in socket chat functions

CI/CD Improvements

None

Technical Details

None

Dependencies

None

Breaking Changes

None

Known Issues

None

v0.12.0

Release v0.12.0

This release introduces a comprehensive Engine and Runtime management system, interactive network visualization, and unified UI/UX improvements.

New Features

• Shared Runtime Library System: Implement .bairuntime packages to reduce package duplication across engines
• RuntimeManager: Full runtime package installation, update, and management
• Environments Page: Comprehensive UI for managing engine and runtime packages with download progress
• EngineManager: Complete engine package management with registry client, downloader, and resolver
• Engine Resolution: Smart fallback chain in ProcessManager (user-selected → default → compatible → bundled)
• First-Run Experience: GPU detection and automatic bundled engine extraction on first launch
• EngineDetailsDrawer: Engine details view with Overview, Files, and Dependencies tabs
• RuntimeDetailsDrawer: Runtime details view with Overview, Files, and Dependent Engines tabs
• BaseCard Component: Reusable card component for consistent UI patterns
• Network Visualization: Interactive network topology view in API Settings using React Flow
• Floating Download Queue: Unified download progress for engine/runtime installations
• Offline Engine Installation: Support for external registry and offline package installation
• HIP Runtime: Windows AMD GPU support with separate HIP runtime packages
• Multi-Accelerator Support: Add SYCL (Intel oneAPI) and Vulkan accelerator support
• Badge Component: Reusable badge with variants (default, primary, success, warning, danger, info)

Improvements

• Upgrade axum 0.7 → 0.8 and tower-http 0.5 → 0.6
• Migrate once_cell::sync::Lazy to std::sync::LazyLock
• Consolidate futures crates (remove futures, keep futures-util)
• Unify engine and runtime registry into single packages.json
• Unified card UI/UX with hover lift effects and loading animations
• Engine/Runtime cards converted to icon-style buttons
• Variant selection dialog improved with descriptions and hardware recommendations

Bug Fixes

• Fix ZIP Slip and path traversal security vulnerabilities in package extraction
• Fix Windows PATH length limit handling (32767 char limit)
• Fix ESLint errors and TypeScript issues across the codebase
• Update failing tests for MarkdownContent, NotificationCenter, and PlatformInfoCard
• Fix runtime manifest library paths to support subdirectories
• Fix semver format for HIP runtime version
• Remove Linux CUDA builds (uses system CUDA instead)
• Fix artifact filtering in publish workflow to prevent cross-contamination

CI/CD Improvements

• Add build-runtime.yml workflow for CUDA and HIP runtime packages
• Add publish-runtime.yml workflow for runtime releases
• Update publish-engine.yml to use unified packages.json
• Remove release trigger from build workflows (independent versioning)

Technical Details

• Engine packages use .baiengine format with manifest v2 supporting runtimeDependencies
• Runtime packages use .bairuntime format with manifest schema
• Packages index (packages.json) combines engines and runtimes for single HTTP request
• Network visualization uses @xyflow/react with custom nodes for User, Router, Backend, and Model

Dependencies

• axum: 0.7 → 0.8
• tower-http: 0.5 → 0.6
• Removed: once_cell, futures
• Added: @xyflow/react for network visualization

Breaking Changes

None

Known Issues

None

What's Changed

• feat: Define .baiengine package format and build system by @inureyes
• feat: Set up engine release repository and index files by @inureyes
• feat: Implement EngineManager module in Rust by @inureyes
• feat: Integrate EngineManager with ProcessManager by @inureyes
• feat: Implement Environments page for engine management by @inureyes
• feat: Implement bundled engine and first-run experience by @inureyes
• fix: Align Environments UI with design system by @inureyes
• fix: Use GitHub Raw URL for engine registry and add local build script by @inureyes
• update: Download release artifacts when target platform differs from current by @inureyes
• feat: Implement offline engine installation with external registry by @inureyes
• fix: Simplify features section subtitle for clearer messaging by @inureyes
• fix: Update Windows and Linux icons in downloads section for better OS representation by @inureyes
• feat: Add interactive network visualization to API Settings by @inureyes
• feat: Implement shared runtime library system to reduce package duplication by @inureyes
• refactor: Cleanup duplicate and unused crate dependencies by @inureyes
• feat: Unified card UI/UX with runtime drawer and animation effects by @inureyes
• docs: Add Engine Management documentation and enhance acceleration guides by @inureyes
• fix: Resolve ESLint errors, TypeScript issues, and update failing tests by @inureyes

v0.11.0

Release Notes for v0.11.0

This release focuses on Agent system enhancements with parallel tool execution, MCP tool integration, new filesystem tools, and Windows compatibility improvements.

New Features

• Parallel tool execution for improved agent performance with concurrent tool calls
• MCP (Model Context Protocol) tool integration with Agent system for external tool extensibility
• run_shell tool with comprehensive security restrictions for shell command execution
• create_directory tool for Agent filesystem operations
• search_content tool for regex-based file content search with context lines support
• Default reasoning effort setting in Settings > Inference tab
• 'Create new folder' option in Move to Folder dialog for inline folder creation
• Enhanced tooltip visibility for tool status indicators in ToolSelector
• OS-specific logos in download section (Apple, Windows, Linux)

Improvements

• Bump Continuum Router version to v0.32.0
• Documentation for new Agent tools (run_shell, create_directory, search_content)
• Documentation for parallel tool execution and MCP integration
• Documentation for reasoning effort feature

Bug Fixes

• Windows compatibility for run_shell tool with proper process tree cleanup using Windows Job Objects
• Windows compatibility for search_content tool with AppData path security
• Windows build errors for Job Object API (Win32_Security feature, HANDLE type conversion)
• Tilde expansion (~) and error message propagation for filesystem tools
• macOS header left padding for mobile view to prevent traffic light overlap
• Text vertical alignment in benchmark result export modal
• Rename icon changed to simple pencil icon
• Korean spelling errors and webpage content improvements

CI/CD Improvements

• None

Technical Details

• Parallel tool execution uses atomic counters and oneshot channels for race-free approval handling
• MCP tools are prefixed with server name (e.g., "mcp:server:tool") to prevent name collisions
• Rate limiting for MCP tool calls (60 calls/minute default per server)
• Path security validation for all MCP tool arguments that look like file paths
• search_content includes ReDoS protection, binary file skipping, and result limiting

Dependencies

• Continuum Router v0.32.0
• windows crate now includes Win32_Security feature for Job Object API

Breaking Changes

• None

Known Issues

• None

v0.10.0

Release v0.10.0

This release introduces a unified Notification Center for better visibility into background processes, API key authentication for llama-server, improved model naming, and numerous bug fixes and documentation improvements.

New Features

• Unified Notification Center for background processes and system alerts
• Global notification system for API errors
• API key authentication for llama-server process
• Humanized display names for all local models in model selector

Improvements

• Bump continuum-router version to v0.31.4
• Massive expansion of Troubleshooting guide
• Documentation content enhancement and structural improvements

Bug Fixes

• Date order of statistical graphs
• Chat deletion popup now respects application theme
• List icon visual balance in display toggle button
• Remove orange color flash on app startup (FOUC issue)
• False warning for CORS origin patterns with wildcard port like 'http://localhost:*'
• Center align display toggle button icon in Available Models card
• Mobile-friendly minimum window dimensions
• Hide console window flash on Windows when spawning system commands
• Race condition preventing image generation model downloads
• Unify component design system across Model Benchmark tabs

CI/CD Improvements

None

Technical Details

• Unified notification system architecture for better user experience
• Improved security with API key authentication for local inference server
• Enhanced cross-platform compatibility with Windows-specific fixes

Dependencies

• continuum-router: v0.31.4

Breaking Changes

None

Known Issues

None

What's Changed

• Docs: Content Enhancement and Structural Improvements by @inureyes
• Docs: Massive expansion of Troubleshooting guide by @inureyes
• fix: Unify component design system across Model Benchmark tabs by @inureyes
• fix: resolve race condition preventing image generation model downloads by @inureyes
• fix: hide console window flash on Windows when spawning system commands by @inureyes
• fix: set mobile-friendly minimum window dimensions by @inureyes
• fix: center align display toggle button icon in Available Models card by @inureyes
• fix: False warning for CORS origin patterns with wildcard port like 'http://localhost:*' by @inureyes
• fix: Remove orange color flash on app startup (FOUC issue) by @inureyes
• feat: add API key authentication for llama-server process by @inureyes
• feat: Use humanized display names for all local models in model selector by @inureyes
• fix: Chat deletion popup now respects application theme by @inureyes
• feat: add global notification system for API errors by @inureyes
• update: bump continuum-router version to v0.31.4 by @inureyes
• feat: add unified Notification Center for background processes and system alerts by @inureyes

v0.9.1

Release v0.9.1

This release focuses on Windows platform fixes and comprehensive documentation updates.

New Features

• Comprehensive user manual with mkdocs documentation system
• Smart download feature and unified manual design with landing page

Improvements

• Update English UI strings to sentence case capitalization
• Update continuum-router to v0.31.3

Bug Fixes

• TCP fallback for Chat and Agent on Windows
• Unix socket config skip on Windows to fix router startup
• SSE streaming responses from cloud backends in agent LLM client
• Title generation for Anthropic and cloud providers via unified router
• ESLint warnings for non-null assertions and console statements
• Unix-only constants with cfg(unix) to suppress Windows warnings
• CI symlink loop and missing doc files

CI/CD Improvements

• Comprehensive manual update with documentation structure improvements
• Documentation override files moved to docs/overrides

Technical Details

• Full compare:

Dependencies

• continuum-router: v0.31.2 → v0.31.3

Breaking Changes

None

Known Issues

None

What's Changed

• docs: update README with new architecture and features by @inureyes
• docs: create comprehensive user manual with mkdocs by @inureyes
• fix: handle SSE streaming responses from cloud backends in agent LLM client by @inureyes
• feat: unify manual design with landing page and enhance smart download by @inureyes
• docs: update manual with latest features and corrections by @inureyes
• fix: skip Unix socket config on Windows to fix router startup by @inureyes
• fix: add TCP fallback for Chat and Agent on Windows by @inureyes
• chore: update continuum-router to v0.31.3 by @inureyes
• fix: title generation for Anthropic and cloud providers via unified router by @inureyes
• fix: Update English UI strings to sentence case capitalization by @inureyes

v0.9.0

Backend.AI GO v0.9.0

This release adds tool permission system integration, MCP tool calling support for external LLM providers, Flatpak distribution for Linux, and various improvements to build security and infrastructure.

New Features

• Tool permission system integrated with AgentLoop execution for controlled tool access
• MCP (Model Context Protocol) tool calling for external LLM providers
• Flatpak distribution support for Linux users
• Download webpage for Backend.AI GO

Improvements

• Build hardening and obfuscation for release builds
• Static export path handling with trailingSlash configuration

Bug Fixes

• Linting and formatting errors across Rust and TypeScript codebase
• Settings icon replaced with proper GearIcon component
• Unix socket Windows changes reverted to restore API provider functionality
• Unused import and constant warnings on Windows

CI/CD Improvements

• GitHub Actions workflow for Next.js deployment
• Webpage workflow improvements for correct project structure

Technical Details

• Integrated tool executor with permission checking from settings
• Tool registry architecture with Arc-wrapped sharing for AgentLoop

Dependencies

• None

Breaking Changes

• None

Known Issues

• None

What's Changed

• feat: integrate tool permission system with AgentLoop execution by @inureyes