Skip to main content

browser

Drive a real browser: open a page, act on it (click, type), read the parts that matter. It holds a session and can change pages, so it is its own capability - distinct from web (search/fetch of the public web).

Exact tools and parameters

catalog.describe_module browser returns the real tools and every parameter, from the live registry.

Tools, at a glance: open, act, read.

Safety switches (config)​

yaml
tools:
  modules:
    browser:
      config:
        actions: none     # "none" = look only, never touch; anything else = acting allowed
        profile: false    # true = use the user's visible login profile
        submit: false     # true = the agent may press a control that posts a form
  • actions is a binary switch: none = look-only; any other value = acting is allowed.
  • Whether a human is asked before an action is not decided here - the app opts into that with tools.capabilities.approve on browser.act, the same approval gate every tool uses.
  • submit is the one commit-shaped action gated separately: off by default, and only the app (never the agent) can turn it on.

A working app​

Compiles as-is. The brain uses a Digitorn gateway model (the default - no API key of your own needed); see the note below to use your own key or a local model instead.

yaml
app:
  app_id: browser_pilot
  name: Browser Pilot
runtime:
  entry_agent: main
agents:
  - id: main
    system_prompt: |
      Drive a real browser to complete web tasks: open a page, act on it
      (click, type), and read the parts that matter.
    brain:
      provider: openai
      backend: openai_compat
      model: mimo-v2.5-free   # a model the Digitorn gateway serves
      config:
        api_key: placeholder  # ignored in gateway mode
tools:
  modules:
    browser:
      config:
        actions: act
  capabilities:
    grant:
      - module: browser
        tools: [open, act, read]