MindMux Desktop
The local-first AI workbench with a project brain.
August 2026
Skills catalog, session personas, and a polish pass
Browse and install curated skills right from the app, see which expert persona is active in a chat, plus fixes to chat titles, Mermaid theming, and settings surfaces.
Read update →Tahoe theme and a post-run panel for Goal mode
A native macOS Tahoe theme arrives, and Goal mode runs now end with a delivery panel for merging or checking out results — plus reliability fixes for chat cards and startup.
Read update →Customize surface, Inbox GA, and split language settings
A new Customize surface unifies Connectors, Skills, and the Artifacts Hub; Inbox is now generally available with GitHub PR acceptance. Interface and AI-reply languages are now separate settings.
Read update →Import .docx documents directly into a project
Projects now accept .docx imports directly, plus fixes for page header dates and selection dialogs.
Read update →Import documents into your brain from chat
Share a file with the AI agent during a conversation and it gets pulled into your project brain — no separate import step. Plus Codex runtime reliability fixes.
Read update →July 2026
Redesigned chat right panel and deck style picker
A redesigned chat right panel, a new style picker for decks, and model names on assistant turns — plus fixes for Mermaid flicker and dropdowns.
Read update →Artifact completion previews and first-class Bento decks
Artifacts preview as work finishes, Bento-format decks get first-class support, and cloud sync recovers gracefully from expired sessions.
Read update →Session approval modes
New session approval modes, with fixes to approval labels and page tag limits.
Read update →Inbox drawer and a Generate menu in the composer
Kick off document, deck, and other output generation right from the composer's new Generate menu — plus an inbox drawer and stability fixes.
Read update →Rosé Pine and Tokyo Night themes, redesigned onboarding
Two new atmosphere themes with unified surface styling, and a first-run setup that drafts your project brain's initial pages for you.
Read update →Faster session switching and a design consistency pass
Switching between long-running sessions is noticeably faster, plus fixes for title bar contrast, status colors, and brain health repair.
Read update →Timeline authorship and more parallel Goal mode
Timeline entries now show authors, Goal mode dispatches independent work in parallel, and the left sidebar collapses across all spaces.
Read update →Reliability fixes for long-running tasks and Brain Health
Long-running tasks recover instead of hanging, and Brain Health repair drafts stay UI-only until you confirm them.
Read update →Brain Health: periodic checks on your assistant's state
MindMux now runs periodic Brain Health checks and surfaces relevant follow-ups — plus fixes for phantom assistant messages and inbox triage.
Read update →Build reliability fix
Fixed an out-of-memory failure in the renderer build by prebundling heavy dependencies with esbuild.
Read update →Work Until Done mode and Cloud Brain sync (beta)
WUD mode keeps working until a task is actually finished, and your brain can now sync to the cloud — for personal use and shared team brains.
Read update →Fixes for task environment variables and packaged document tools
Environment variables from settings.json now apply to built-in Claude task runs, and document tools unpack correctly in packaged builds.
Read update →Artifacts can now revise previously generated files
Iterate on generated content without starting from scratch, plus a fix for window layout persistence and a UI polish pass.
Read update →@-mention resources from the composer
Type @ in the composer to quickly add resources, plus a tidier inbox button and visual refinements across the app.
Read update →More precise multi-keyword search
Search now matches pages containing all of your terms, plus a permission-gated write channel for the assistant and several chat fixes.
Read update →Deep Research tasks
A new Deep Research task type runs bounded parallel search waves and compiles a structured report — and the assistant can now cancel tasks mid-run.
Read update →Interface rebuilt on shadcn/ui and Tailwind
The app's interface has been rebuilt on shadcn/ui and Tailwind for a more consistent look and feel, plus fixes for workspace skills and chat cards.
Read update →June 2026
Long-running operations no longer block the UI
Cards that previously hung indefinitely are now suspended and resumed in the background via a ticket system, keeping the app responsive.
Read update →Sidebar alignment fix and shadcn styling refresh
Fixed session action button alignment in the sidebar, and refreshed UI styling to align with shadcn component defaults.
Read update →Fix hardcoded colors ignoring the active theme
Fixed a P0 bug where UI elements rendered with hardcoded colors instead of respecting the active theme.
Read update →Configuration Expert handoff in chat
The assistant can now hand off to a dedicated configuration expert, and the packaged app now respects proxy environment variables.
Read update →Global search and Max reasoning tier
Search across all chats and pages from anywhere in the app, and access Anthropic's top reasoning tier — plus fixes for zoom, branch layout, and draft protection.
Read update →Windows build (early access) and brain root in onboarding
MindMux now ships a Windows installer with platform-specific polish, and you can pick your brain root during onboarding.
Read update →Fix for first messages disappearing after session start
Your opening message is now preserved when a session starts, and dropdown menus no longer clip or misalign.
Read update →Redesigned Brain navigation and OpenAI auth for Codex
Refreshed Brain navigation, OpenAI credentials for Codex providers, and a fix for connectors failing in packaged builds.
Read update →Custom providers in onboarding and AI-generated pages
Set up a custom AI provider during onboarding, pages are now created by the AI, and a new diagnostics view verifies your provider connection.
Read update →Page lifecycle status; Library retired
Pages now carry an active/archived status that controls AI context inclusion, and the Library feature folds into Draft Pages.
Read update →Fix for title bar hidden behind other UI
The title bar now reliably stays on top, keeping window controls accessible.
Read update →Chat header now shows the active expert
See which expert you're speaking with at a glance, plus fixes for dropdown overlap and input alignment.
Read update →Fix skill imports via symlinked paths
Skills stored or deployed through symlinked paths now load correctly again.
Read update →Open Recent projects and quicker page management
Jump back into recent projects from the app menu, reach Settings and update checks directly, and delete brain pages from the context menu.
Read update →Desktop notifications and directory-based skill bundles
Stay in the loop with desktop notifications for chat messages and interactive requests, and package skills as directories for richer setups.
Read update →AI inbox triage, artifact tasks, and a redesigned chat layout
The inbox now uses AI to triage incoming items, artifacts get a dedicated task type, and the chat view gets a refreshed top-nav layout.
Read update →Agents can run shell commands, with an allowlist
Agents can now execute shell commands from chat under a configurable allowlist, plus a sidebar model selector and a Help menu with feedback.
Read update →Chat file attachments and Onboarding v2
Attach files to your very first chat message, a redesigned two-step onboarding, and smoother runtime switching.
Read update →Context preserved across runtime switches
Conversation context survives runtime changes, the Claude runtime fetches its model list live, and Copilot falls back to a bundled CLI.
Read update →May 2026
Flatter, easier-to-scan sidebar
Sessions now display in a single-level list instead of a nested hierarchy, making the sidebar easier to scan and switch between.
Read update →Fix for Gatekeeper warning on fresh macOS installs
New downloads install cleanly again — a packaging problem that dropped the Electron Framework and triggered a Gatekeeper warning is fixed.
Read update →