Version history
See what is new, improved, and fixed in every Luxi release.
Optional GenUI brings interactive charts, tables, and forms to replies. Reports and documents default to HTML files when no format is specified, and user messages now appear before their working and compaction status.
Restores visible character counts and elapsed time during reasoning and fixes an object-destroyed error when closing the main window.
Share web pages with AI inside Luxi and pause to take over, with updates to the conversation runtime, context display and MLX inference engine.
Improve model download recovery, adjust the default reasoning budget to 8K tokens, and update local inference compatibility with UTF-8 output guarding.
Updated Pro/Lite local models, startup warmup, a default 12K-token reasoning budget and simpler settings. Download MLX models when upgrading; existing data is preserved.
Search and highlight matches within chats, navigate history, and check context capacity before switching models.
Background model initialization and explicit retries, model-file repair, sidebar conversation deletion, and clearer errors.
Improves startup and update feedback, model selection, and saving reliability; adds drag-to-collapse sidebars and isolates Luxi from standalone DSH.
Updated the sidebar workbench, added local Office previews and browser-opening actions, and improved scheduled task management and conversation reading.
Bundled a Python document runtime and Chinese fonts, with improved guidance for document and data tasks, workspace dependency management, and local model recovery.
Added collapsible reasoning and tool progress, and improved local model startup, validation retention, and long-conversation handling for clearer waits and smoother continued use.
Adds one-click download, resumable transfers, integrity verification, and automatic validation for the local StartLux model, while fixing model availability after restart and unwanted navigation when activating the app.
Continue sessions after long responses are truncated, opt out of privacy-safe anonymous error reports, and validate local StartLux Qwen GGUF models before using them.
The sidebar update entry now shows live download progress, produced files include what your scripts write, and the copy buttons work again.
The composer "+" menu is rebuilt around four labeled entries, and the DuckDuckGo search fallback is now visible in settings and reaches the network through your system proxy by default.
Adds archived conversations management in settings and version-stamps release artifacts to avoid stale CDN caches.
Shows the current app version with a sidebar update button, caps Luxi Lite context at 64K, and fixes clean-install and release pipeline issues.
Adds scheduled task automation and in-app auto-update, a DuckDuckGo web search fallback, and makes a just-downloaded model ready to use immediately.
Fixes local-model tool calling, offline Agent usage, and web-search reliability
Projects, sessions, and document artifacts now share a coherent workspace, with stronger Agent execution and local-model reliability.
Local models can now understand image input, with more reliable PDF extraction dependency loading in the desktop app.
More reliable Office/PPTX tasks, custom local models, persistent agent context, and improved feedback.
Faster, more resilient local-model downloads and a smoother first-run experience for Apple Silicon Macs.
The first Luxi release with local inference, anonymous product feedback, and in-app updates for Apple Silicon Macs.