Everything Armadai does.
In one place.

A new canvas starts with a setup card. Pick an agent that’s ready, or follow Install → Sign in → Ready; Set up… installs it, signs you in, and makes it the master.

Type, dictate, or attach. Drag in files, folders, and images, or paste a screenshot; HEIC and large images are converted for you.

Keep sending; your messages wait their turn. Stop the master’s turn without stopping its agents, and read long runs folded into one line.

Type / for a ranked menu of Armadai’s commands, each agent’s own commands, your skills, and MCP prompts. /claude opus high switches agent, model, and effort in one step.

Switch modes with Shift+Tab or /mode. When a plan is ready, a card asks whether to implement it, and No, keep planning opens a box for what to change.

Bring up each agent’s own permissions screen inside Armadai: Claude Code’s rules and tabs, Codex’s pick list, or Ask and Always approve for Grok and Cursor.

Ask the master to report back when an agent finishes. It tells finished work, plans ready for review, and blockers apart, and you can cancel a follow-up without stopping the agent.

Connect agents through explicit links with roles, messages, status, transcripts, and approval handling. The master delegates work and collects the results.

Short requests such as “open the shop workspace” happen instantly: show a workspace, start an agent, tidy up, lay out the canvas, or add a browser.

Keep any number of persistent canvases. Pan, zoom, move, resize, and layer tiles; each canvas keeps its own theme, layout, and master conversation.

Canvas View shows the whole spatial layout; Tab View shows one workspace at a time. Dragging a tile brings it and its workspace forward.

Add workspace → New workspace… creates a project folder, and Open existing folder… brings in one you have. Arrange them freely, in a row, a column, or a grid.

Shell terminals, Claude Code, Codex, Grok, Cursor, Codex Agent, a VS Code–based editor, browsers, notes, task boards, visualizations, and iOS simulators.

Switch to a compact floating dock without restarting agents. It holds masters, canvases, sessions, approvals, and scheduled tasks, and remembers its place across displays.

A menu bar item opens the master, canvases, agents and approvals, and scheduled tasks. Show Master has a shortcut: ⌃⌥A or ⌃⌥Space.

Pure Black, Graphite, Deep Ocean, and Dusk Glass. Each canvas can have its own theme and transparency.

A ⌘K command palette, shortcuts, context menus, and settings search. Two fingers pan the canvas; hold Option to scroll inside a tile.

Keep notes beside your work and draw on canvases or workspaces. The master draws editable flowcharts, sequences, trees, and grids, and revises them on request.

A VS Code–based editor sits beside shell terminals and agents. Run dev servers, builds, and tests, and drop files on a terminal to paste their paths.

A structured Codex conversation with streamed replies, tool activity, and approvals. Copy or Fork any reply, into this workspace or a new Git worktree.

Give every agent its own worktree, or choose per launch. Inspect agent worktrees, and choose what to keep or remove when an agent closes.

Board, Backlog, Table, Calendar, and Dashboard views of the same work, with priorities, owners, dependencies, checklists, sprints, and work-in-progress limits.

Hand a ticket to a coding agent in its own worktree and follow it through to review. GitHub workflows prepare pull requests linked to the task and its evidence.

Ask an agent to build charts, simulations, or dashboards as HTML or React tiles. Preview them, publish a working version, view the source, or export the project.

Run a simulator inside a workspace: choose the device, tap and type, rotate, take screenshots, and build and run the workspace’s iOS app.

Tabs with pinning, find on page, per-tab zoom, and downloads, and search with Google, DuckDuckGo, Bing, or Brave. The master can read console errors and capture pages.

Review working or staged changes, stage files, commit, and confirm a push, all from the Codex Agent tile.

Set instructions, canvas, workspace, agent, model, effort, account, and approvals: Automatic or Wait for me. Repeat hourly, daily, on weekdays, on chosen days, or by cron.

Pause, edit, or Run now. Each run keeps its results, what the agent reported, evidence, and the conversation, and you can message the task’s agent directly.

At a usage limit, a run tries your other accounts, then other agents in the order you choose. Interrupted runs pick up in the same tile, worktree, and conversation.

Add and name more accounts, then choose one for each master, workspace, tile, or task. Usage shows in Accounts, the sidebar, the master’s header, and agent tiles.

When an account runs out, continue on another account in the same conversation, hand the work to another agent with a summary, or wait for reset. Automatic switching counts down from 5 seconds, with Undo.

Save keys for Anthropic, OpenAI, xAI, and Cursor. Bill a subscription, an API key, or a subscription with API-key fallback, for the whole app or one master, workspace, agent, or task.

Canvases, layouts, and conversations are saved. The session helper keeps terminals running while the app is closed, and agents resume after restarts and updates.

Optionally record terminal output on your Mac. Browse, export, and delete recordings, and set how long they’re kept and how much space they may use.

Update Armadai and your agent CLIs together, timed around agent activity. Agents resume their conversations afterwards.

Masters and agents read and operate Mac apps through accessibility information, screenshots you allow, and mouse and keyboard actions.

Turn computer use on for each canvas and each agent. Shared resources are coordinated between agents, with foreground-takeover rules and Stop controls everywhere.

When a step in Chrome needs more than the extension, such as a native dialog or a browser shortcut, Armadai completes it with computer use.

An optional fast assistant. It carries out quick requests instantly, reads unfamiliar agent screens, and tells when an agent has finished.

With Let Jev see screenshots on, Jev finds unlabelled controls during computer use. In a simulator, agents can tap a control just by describing it.

With Jev on, Claude Code’s Auto setting picks Sonnet or Opus for each message.

See which agents are working, waiting, blocked, or asking, across tiles, the master, the sidebar, the dock, and your iPhone. Hover a waiting agent to answer with its real choices.

Put each agent on Auto or Manual approval, or turn on Auto-approve all agent requests with Except what Jev flags to leave risky requests to you.

Chime, Whoosh, Beep, or Pop when an agent finishes or needs approval, with volume and a preview.

Logins, cards, secure notes, recovery codes, and custom fields in an encrypted vault on your Mac, with search, categories, favorites, tags, and archives.

Each item says which agents may find or use it. Protected login fills credentials for them, card fills ask you first, and sharing a note’s content is a separate choice.

Unlock with your password or Touch ID, auto-lock after 10 minutes, 1 hour, or 6 hours, or press ⌘⇧L. Authenticator codes, Chrome import, and encrypted backups are built in.

Keep environment variables and secrets in the macOS Keychain, scoped to the app, a canvas, or a workspace. Import .env files; newly launched agents get the values.

Sign in with an emailed code, email and password, Apple, Google, or X. Connect several methods to one account, and change your email or password anytime.

Turn on Use Touch ID on this Mac and Armadai locks at launch until you unlock it.