Integrated browser

The editor has an integrated browser, driven by Playwright in-process — no separate Chrome or Chromium to install. Its tools are gated behind workbench.browser.enableChatTools, which Sirius turns on by default (upstream ships it off). With the gate on and agent mode enabled, the Sirius agent offers six browser tools in its extended tier:

Tool	What the model is told
`open_browser_page`	Open a new browser page in the integrated browser at the given URL; returns a page id for the other tools
`read_page`	Get a snapshot of the current page state — the accessibility tree; “better than screenshot”
`screenshot_page`	Capture a screenshot of the current page; for looking, not for acting
`navigate_page`	Navigate by URL, history or reload
`click_element`	Click an element
`type_in_page`	Type text or press keys

read_page is the primary path: it returns structured text any model can use, including small local ones. screenshot_page returns an image, which reaches the models that accept images — Claude, Gemini, vision-capable Ollama models and the OpenAI-compatible vision models Sirius recognises; for others the image is replaced by a note that the model cannot view it (Image input).

The extended tier is offered to hosted models, local models of 4 GB or more, and unknown-size models with a context window of at least 32,000 tokens; a small local model gets the core tools only, so it is not handed sixteen schemas it cannot hold.

What is not offered

The editor registers a few more browser tools — hover_element, drag_element, handle_dialog, run_playwright_code — that the Sirius agent does not pass to the model in 1.118.7. Driving your own Chrome instead of the integrated browser is not implemented.

Turning it off

Set workbench.browser.enableChatTools to false. With the gate off the agent can still open a page for you, but has no access to its contents. There are no sirius.ai.browser.* settings; the tools are the workbench’s.