The desktop app

The desktop app is a native app for macOS, Windows and Linux that carries mu and its runtime. It runs the same agent and judgment kernel as the command line, in a process of its own, so everything the command line can do, the app can do. It adds a work panel next to the conversation, so you can see what the agent and the judge are doing.

The conversation

One list for both. The conversation list is mu's own session files. A session you started in a terminal shows up in the app, and one started in the app can be continued with mu -c or mu -r.
Quiet by default. What mu says to the model on its own, such as a reminder to verify, a goal continuation, a new error or a rewind, shows as a single line in your language; the original text is one click away. A run of tool calls is folded into one line that says what was done, such as read 3 files · ran 1 command.
Fork from any message. Each message has a fork button that starts a new branch from that point.
Row menu. Rename, pin, mark as unread, archive or delete a conversation.
Notifications. When another conversation finishes or waits for you, the system tells you.

The send box

Permission mode. Full access, Jev approves or minimal permissions, switched per conversation. See Permissions and safety.
Goal. Set a goal, and the agent keeps working until it holds. While a goal runs, a goal line above the box shows its state. See Goal mode.
Context ring. How full the context is, at a glance.
Queued messages. A message sent while the agent works waits in a queue and can be taken back before it is used.
Slash commands. Type / for mu's and pi's commands, the same as in the terminal.

⌘K (Ctrl K on Windows and Linux) opens the command palette.

The work panel

The panel beside the conversation has eight tabs:

Tab	What it shows
Board	The plain-language board: how far the work is, what is happening now, what waits for you, and a running account
Judgments	The live ledger: every verdict with its question, its answer, its probability and how long it took
Hive	The hive: each bee, what it is doing, and a map of which finding went to whom
Lessons	The lessons library: what mu has learned, how often each lesson was recalled and followed; edit or retire one
Files	The project's files
Preview	A preview of the page or file being worked on
Source	The folder's git changes and their diffs
Browser	The built-in browser the agent drives step by step, with its goal; pause it, take over, or stop it

The built-in browser keeps its own profile and never touches your own browser. Each step shows what was clicked and how sure the judge was; submitting, paying or deleting stops and asks you first.

Signing in to models

Settings → Models signs you in without leaving the app: a ChatGPT, Claude, Grok or Google (Gemini CLI or Antigravity) subscription, or an API key for any provider pi supports. A server of your own works too, including a plain http:// address on your machine or network. The thinking level of each model can be changed there.

Google's two sign-ins are Google's logins for its own tools. Using them from another program may break Google's terms, so the app states the risk first and goes on only if you agree. A Gemini API key and Vertex AI need none of this.

Settings

Settings are grouped by area, each page short: Preferences, Models, Kernel, Details, Capabilities and System. A search box at the top of the sidebar finds any setting.

Kernel lists each feature as one row with a switch; its decision points are listed under it, each with its own switch. The judges page chooses who answers.
Capabilities lists skills by source (mu's own, the project's, or inherited from Claude Code, Codex and Cursor), lets you add a skill folder and remove your own; and lists MCP servers, each with a switch. Keys passed to a server as arguments are masked.
Everything the settings change is written to ~/.mu/agent/mu.json, so it holds for the command line too, and the other way round.

Light and dark follow the system. The interface comes in 13 languages, including Simplified and Traditional Chinese, English, Japanese and Korean.

Importing conversations

Conversations from Claude Code and the Codex CLI can be imported and continued in mu. The first-run guide offers this when it finds some; the command line does it with mu import --list and mu import <file>.

Updates, proxies and logs

Updates. The app tells you when a new release is out, with its size; it downloads only after you agree and installs at the next restart.
Proxy. With no HTTPS_PROXY set, mu, the sign-ins and the key checks in the app follow the system proxy.
Logs. Settings → About opens the log folder.