Skip to main content

Agent tools

Tools are the agent's hands. When the agent decides it needs to do something — fetch a URL, send a message, query Home Assistant — it calls a tool. Tools have well-defined inputs and outputs that the LLM can reason about.

Built-in tools​

Out of the box you get tools for:

  • Web — browse a URL, search the web
  • Filesystem — read, write, list, search files
  • Bash — run shell commands (with optional confirm-before-run)
  • Memory — save, recall, search semantic memories
  • Channels — send and read messages on connected integrations (Discord, Slack, Telegram, WhatsApp)
  • Image generation — DALL-E, Stable Diffusion, and other providers selectable from the registry
  • Voice — synthesize speech, transcribe audio
  • Calendar / mail / contacts — Google and iCloud, when those integrations are configured
  • Plan / schedule / spawn — split work into a plan, schedule a future call, spawn a sub-agent

The full reference lives in the developer docs (coming).

Per-agent toolsets​

Each agent has a whitelist of tools it can use. By default, restricted to what makes sense for that role. You can edit the whitelist from the agent's settings page.

Confirming dangerous actions​

Some tools (delete files, send to a contact) require explicit confirmation before running. Configurable per-tool.

Custom tools​

To add your own tools, package them in a plugin — see the plugin authoring guide.