Command Reference
Canonical public command reference for Comet Browser, aligned with the landing page AI guide. Commands are documented JSON-first for reliable execution and indexing.
AI Guide Note
This page follows the public command set documented in Landing_Page/AI-GUIDE.md. Apple Intelligence stays documented under the native API, and cross-app OCR/click flows are described as native-first with OCR fallback.
Risk Levels:
NAVIGATE
Bracket/JSON Format
Navigate to a specific URL. Supports HTTP, HTTPS, and file:// URLs.
Parameters (1 required)
urlSEARCH
Bracket Format
Perform a web search using the default search engine.
Parameters (1 required)
queryWEB_SEARCH
Bracket Format
Perform a real-time web search with RAG (Retrieval Augmented Generation) for up-to-date information.
Parameters (1 required)
queryRELOAD
Bracket Format
Refresh the current browser tab.
Parameters (0 required)
GO_BACK
Bracket Format
Navigate back in browser history.
Parameters (0 required)
GO_FORWARD
Bracket Format
Navigate forward in browser history.
Parameters (0 required)
READ_PAGE_CONTENT
Bracket Format
Extract and read text content from the active browser tab.
Parameters (0 required)
LIST_OPEN_TABS
Bracket Format
List all currently open browser tabs with their titles and URLs.
Parameters (0 required)
CREATE_PDF_JSON
JSON Format
Generate a structured, multi-page PDF document with advanced formatting options. PREFERRED method.
Parameters (2 required)
titletemplatecontentbranding+1 moreCREATE_FILE_JSON
JSON Format
Create various file formats (PDF, PPTX, DOCX) using structured JSON. Supports rich media slides.
Parameters (3 required)
formattitleslidesSHELL_COMMAND
JSON Format
Execute terminal commands on the host operating system. All shell commands require explicit user approval.
Parameters (1 required)
commandshellSET_VOLUME
Bracket Format
Adjust the system volume level.
Parameters (1 required)
levelSET_BRIGHTNESS
Bracket Format
Adjust the display brightness level.
Parameters (1 required)
levelOPEN_APP
Bracket Format
Launch an application on the host system.
Parameters (1 required)
appurlOCR_SCREEN
JSON Format
Capture the screen or a region and extract text. Comet uses native OS OCR and accessibility providers first, with image OCR as a fallback path.
Parameters (0 required)
regionOCR_COORDINATES
JSON Format
Resolve coordinates for visible text in desktop apps so Comet can target external UI reliably.
Parameters (1 required)
textthresholdCLICK_APP_ELEMENT
JSON Format
Click on an element in an external application using native-first visual matching and coordinate resolution.
Parameters (2 required)
apptextxyLIST_AUTOMATIONS
Bracket Format
List all scheduled automation tasks with their status and schedules.
Parameters (0 required)
DELETE_AUTOMATION
Bracket Format
Delete an automation task using its ID.
Parameters (1 required)
taskIdSCHEDULE_TASK
JSON Format
Schedule a recurring automation task that runs through the background service even when the browser is closed.
Parameters (3 required)
scheduletypenameTHINK
JSON Format
Expose the model's reasoning or intermediate thinking to the user for complex tasks.
Parameters (1 required)
reasoningPLAN
JSON Format
Show the intended execution steps for a multi-step task before or during execution.
Parameters (1 required)
stepsPLUGIN_COMMAND
Bracket Format
Execute a command defined by a plugin. Format: pluginId.commandName | params.
Parameters (1 required)
pluginCommandparamsCLICK_ELEMENT
JSON/Bracket Format
Click on a visible DOM element using visual intelligence (OCR + coordinate mapping).
Parameters (1 required)
textmatchTypeindexFIND_AND_CLICK
JSON/Bracket Format
Find text on the current page and click nearby elements. Uses OCR for text detection.
Parameters (1 required)
findTextclickTextFILL_FORM
JSON/Bracket Format
Fill in form fields identified by label, placeholder, or name.
Parameters (2 required)
fieldsfields[].labelfields[].namefields[].valueQuick Command Reference
NAVIGATESEARCHWEB_SEARCHRELOADGO_BACKGO_FORWARDREAD_PAGE_CONTENTLIST_OPEN_TABSCREATE_PDF_JSONUnderstand the Security Model
Learn how Comet AI protects your system with visual sandboxing, syntactic firewalls, and human-in-the-loop authorization.