Docs

Modes

The launcher is one window with many modes. You can switch modes with a tag at the start of your prompt or a keyboard shortcut, and modes compose — agent + search + vision can all be on at once.

Chat (default)

A direct conversation with a model. No tools, fast responses. Best for quick questions, drafting, and explanations.

  • Trigger: the default when no mode is set.

Agent

Unlocks tool use: the agent can browse the web, read and write files, run code and shell commands, control Chrome, manage Calendar and Mail, and call any connected tools and MCP servers. Sensitive actions require your approval.

  • Trigger: /agent (also /task) or ⌘1.

Kernel

A Python execution environment with persistent session variables, available inside agent mode. Useful for data wrangling and chained computations where state carries between steps.

  • Trigger: ⌘7 while agent mode is active. Can be enabled by default in settings.

Vision

Lets the model see images and screenshots you attach. Auto-enables after you capture your screen.

Search

Runs live web searches and adaptively refines its queries to gather current information.

  • Trigger: /search (optionally /search your query) or ⌘5.

Imagine

Generates images from a text prompt using image-generation models.

  • Trigger: /imagine your prompt or ⌘6.

Think

Asks reasoning-capable models to show extended thinking before answering — good for hard problems where you want the model to work through steps.

  • Trigger: /think (also /thinking). Note: the default global screenshot shortcut also uses ⌘4, so use the tag to be unambiguous.

Bundle (compare)

Runs your prompt against multiple models at once and shows the responses side by side, so you can compare answers or pick the best one.

  • Trigger: ⌘2. Choose which models to compare with the bundle default in Models & Providers (e.g. agentastic:frontier).

Combining modes

Modes stack. For example:

/agent /search Find this week's top AI papers and save the titles
and abstracts to ~/Desktop/papers.md

This runs an autonomous agent with web search enabled. Add /model:opus to also pin the model for the request.

Sub-agents (apps)

Some specialized workflows run as apps launched with /app::

  • Deep Research — /app:deep-research opens a planner, then runs an iterative research pipeline and produces a report.
  • Design — /app:design opens an AI design canvas for visuals, decks, and presentations.
  • Dictation — /app:dictation is the voice-to-text workflow (also triggered by the Fn key).

See Apps and Voice & Dictation.

Reference