Tools

Built-in tools

Every tool the platform ships — 39 of them, in 12 groups. Each group has a page with what the tools do, how an agent gets them, and their exact arguments.

Group	Tools
Web & browser	`web_search`, `web_fetch`, `browser_use`
Knowledge & memory	`search_knowledge`, `remember`, `recall`, `remember_for_workspace`
Code & skills	`code_execution`, `load_skill`, `run_skill_script`, `propose_skill`
Documents & media	`write_canvas`, `generate_document`, `generate_image`, `qr_code`
Charts & interaction	`render_chart`, `render_ui`, `ask_user`
Utilities	`current_time`, `calculate`, `convert_units`
Live data	`get_weather`, `weather_forecast`, `currency_convert`, `market_data`
Wake-ups & fan-out	`schedule_followup`, `map_over`
Travel	`travel_search`, `travel_book`, `travel_trips`, `travel_cancel`
Service desk & avatar	`create_ticket`, `escalate_to_owner`
Research evidence	`record_note`, `record_source`
Agent authoring	`read_definition`, `write_definition`, `validate_definition`, `search_catalog`

Two ways an agent gets a tool

Declared. Listed in the agent's tools: in its Agent.md — web, documents, live data, utilities and the domain tools.
Added by the runtime. Some tools follow from how the agent is configured or where it runs, and are never declared: load_skill when it has skills, remember/recall when it has memory, search_knowledge when it has knowledge, write_canvas and schedule_followup on chat and automation runs, render_ui and render_chart when a person is watching in the web app. Listing one anyway is accepted.

Every tool call goes through the same governance whichever way the tool arrived: tool allowlists, DLP on inputs and outputs, approvals for writes, and the run's budget. Each call is on the run's trace in Observe.

Not on these pages Connector tools (ServiceNow, Google, Microsoft 365…) act as the signed-in person — see Connectors. MCP tools come from servers registered in the Tools Library — see Tools. In the app, Catalog → Tools lists the built-ins alongside both.