Using the Desktop

The Imaginne Desktop is the native app (macOS and Windows) where the agent runs locally on your machine, with access to your files. This page covers day-to-day use. To install, see Install Desktop; for the concepts behind it (sessions, memory, autonomy), follow the links throughout the text.

The layout

The window has three regions:

• Sidebar (left) — the list of Projects with the + button; the workspace navigation (Preview/Diff/Plan, Project Memory, Saved Prompts, Integrations, Run Details, and the command palette ⌘/Ctrl+K); and the Files panel.
• Center — the session header, the conversation transcript, the permission dock (when the agent asks for confirmation), and the message field (composer). The agent's messages appear as Imaginne √-1.
• Right dock — Preview, Diff, and Plan. Full-screen overlays open on top when needed.

Projects and sessions

A project is a working folder on your computer.

• Create — click + in the sidebar. The project is created and a session starts automatically.
• Switch — click another project to change the active workspace.
• Delete — removes the project from the app's catalog, but does not delete the folder on disk.

Each project can have multiple sessions. Use the session menu (in the header) to create, rename, or archive. You can't switch sessions while the agent is responding — cancel or wait. See Sessions.

Chat and slash commands

Type in the composer and send. When the message starts with /, a command dropdown appears.

Action	Key
Send	Enter
Line break	Shift+Enter
Cancel streaming	Esc

Available commands:

Command	What it does
/login · /logout · /whoami	Sign in, sign out, and view your identity (without exposing the token).
/model	Opens the model picker.
/memory	Opens Project Memory.
/clear	Clears the current conversation.
/canvas	Opens/creates a whiteboard.
/image	Generates an image from a prompt.
/publish-skill	Publishes a local skill to your organization.
/fb	Sends feedback to the team.
/help	Lists the commands.

Mode commands (Developer profile)

If your Role is Developer (Settings → Info), you also have /plan, /exe, and /ask to switch the agent's working mode. On Non-Developer accounts these commands don't appear.

Mentioning files: @ and "+ Context"

To point to a project file, type @ followed by the path — this inserts a literal reference to the file in the message.

To include the file's contents in the agent's context, use the + Context button (limit of about 100 KB per file). It's the difference between saying "look at this file" and "here's what's inside it".

tip

Unlike the TUI, the Desktop composer has no @ autocomplete. Type the path or use + Context to attach the contents. See Files & artifacts.

Model

The default model is nnumbers. Open /model (or the picker in the header) to choose another — the list is filtered by your organization's policy. In a single-model organization, the only effective option is nnumbers. See Models.

Autonomy

Autonomy defines how much the agent asks before acting. Adjust it in Settings → Info → Autonomy:

• Low — confirms every action with an effect.
• Medium (default) — approves the trivial and asks for confirmation on the sensitive.
• High — acts without asking (shows a warning when enabled).

The vault (workspace firewall) and the fixed denials stay in effect at any level. See Autonomy & permissions.

Preview, Diff, and Plan

The right dock has three tabs, switchable with Alt+1/2/3:

• Preview — opens the selected file rendered (PDF, DOCX, XLSX, Mermaid). Buttons to open with the OS, @ Insert (inserts a reference in the composer), and + Context (injects the contents).
• Diff — shows the edits the agent made to files.
• Plan — displays the agent's plan (read-only in this version; the Approve/Edit/Reject buttons are still disabled placeholders).

The workspace navigation also has Alt+4/5/6/7 for Memory / Saved Prompts / Integrations / Run Details.

Files

The Files panel shows the project tree. You can:

• New file (Text, Markdown, Word, Excel, or PowerPoint), New folder, delete (✕), and move (drag and drop);
• open with the OS app;
• toggle show hidden.

Beyond the workspace, the agent can read (read-only) from Documents, Downloads, and Desktop, and from iCloud/OneDrive. Writes always go to the workspace.

Generated files are organized automatically by type within the project — images (/image), canvases, and skill outputs.

Voice

There's a microphone button in the composer (hidden when voice isn't available). Click to start and click again to stop. The partial transcription appears in the message field as you speak — and is not sent on its own: you review and send. Voice is speech→text and requires a microphone and credentials enabled by your organization. See Voice.

Saved Prompts and Routines

These are two similar features with different purposes:

• Saved Prompts — reusable prompts per project, run on demand only (Run Now). They have no scheduling, by design.
• Routines — recurring tasks global to the user, with a real scheduler that fires while the app is open. Create them from the Routines drawer → + New routine (fields for title, prompt, schedule, run mode, notifications, and more).

For the full routine configuration and the distinction between the two, see Routines.

Canvas

The canvas is a whiteboard for organizing ideas visually:

• /canvas <title> creates a board;
• /canvas list lists existing boards;
• double-click adds cards;
• ⌘/Ctrl+Z undoes.

Boards are saved in the project.

Presentations

No command to memorize: describe the presentation and the agent builds the deck with the native ppt_skill.

Create an executive presentation about the marketing plan for the second half of the year, 8 slides.

The native engine picks the visual automatically and generates the .pptx (plus PDF and thumbnails when the visualizer is available) as an artifact — you don't choose a template. For the full recipe, see Generate a presentation.

Images: /image

The Desktop generates images from prompts — describe the image and the agent creates it:

/image a friendly robot presenting a chart, flat style

The result appears among the project's artifacts. Press Esc to cancel the generation.

Skills

Skills are capabilities the agent uses when relevant (creating .pptx, .docx, .xlsx, PDF, and so on).

• The bundled skills (docx, excel, pdf, ppt_skill) are installed automatically.
• Your organization's skills are synced when the app starts.
• All of them appear in the / dropdown.

In Settings → Skills you see the bundled ones. On the Desktop, remote install/uninstall is disabled — these skills are embedded. See Skills and the Skills section.

Publish a skill

Use /publish-skill (or the corresponding modal):

1. Choose a local skill.
2. Preview the synthesized manifest.
3. Select a group (required).
4. Click Publish.

Afterward, use Open in /app to manage it in the console. See Publish skills.

Project Memory

Project Memory keeps four persistent notes — project, user, glossary, and decisions — that the agent consults across sessions. You can also use "Pin selected text" to pin relevant snippets to the current session. See Memory.

Remote session

Drive your local session from another screen. In the Remote pill, click Create pair to generate a code (Crockford format) with a 30s countdown and a URL to share (<host>/remote). Buttons for Copy code/URL and Revoke. There's one active pair at a time, and it requires login. See Remote session.

Notifications

The Notifs pill opens the notification center. You get a native system toast (with an in-app fallback) and can adjust preferences by type, plus a Do Not Disturb window. The types include response ready, confirmation request (ask user), tool error, routine completed/failed, remote and skill events, session expired, and update available. See Notifications.

Settings

Open Settings to adjust:

• Info — Name and Organization (read-only), Role (Non-Developer / Developer), Theme (Dark/Light/Auto), Language (EN/PT), Show hidden, and Autonomy.
• Plugins — optional features downloaded on demand (Advanced Preview with LibreOffice, Mermaid, CAD Viewer, Canvas).
• Skills — the bundled skills.
• About — app version and information.

Diagnostics and feedback

If something goes wrong, use Export Support Bundle to generate a redacted diagnostic package (no tokens) and /fb <message> to send feedback to the team. See Troubleshooting.

Keyboard shortcuts

Shortcut	Action
Enter	Send message
Shift+Enter	Line break
Esc	Cancel streaming / close / cancel /image
⌘/Ctrl+K	Command palette
Alt+1 / 2 / 3	Preview / Diff / Plan
Alt+4 / 5 / 6 / 7	Memory / Saved Prompts / Integrations / Run Details
⌘/Ctrl+Z	Undo (in the canvas)
⌘/Ctrl+Enter	Send reply on ask-user

The full reference is in Desktop shortcuts.

See also

⏰Routines

Saved Prompts vs Routines, scheduling, and notifications.

Open →🔗Remote session

Drive your local session from another screen.

Open →🔔Notifications

The notification center and preferences.

Open →🎤Voice

Dictate requests by speech→text.

Open →🧰Skills

What the agent can do and how to publish new skills.

Open →⌨️Desktop shortcuts

The full table of shortcuts and commands.

Open →