browser — drive a real browser session

browser is a single tool with 16 actions for end-to-end browser automation. It wraps Playwright; install with pip install chimera-run[browser].

Schema

Arg	Type	Description
action	string	One of: navigate, click, type, screenshot, content, html, evaluate, wait, select, scroll, back, forward, tabs, new_tab, close_tab, switch_tab.
url	string	For navigate / new_tab.
selector	string	CSS selector for click / type / wait / select.
text	string	Text to type.
js	string	JavaScript expression for evaluate.
path	string	Output path for screenshot.
full_page	boolean	Capture the whole page (default false).
direction	string	up / down for scroll.
amount	number	Scroll pixels.
timeout	number	Action timeout in ms.
value	string	Value for select.

action is required; the other fields are action-specific.

Example invocation

{"action": "navigate", "url": "https://example.com"}
{"action": "screenshot", "path": "/tmp/snap.png", "full_page": true}
{"action": "evaluate", "js": "document.title"}

from chimera.tools.browser import BrowserTool

browser = BrowserTool()
browser.execute({"action": "navigate", "url": "https://example.com"}, env=None)
browser.execute({"action": "click", "selector": "a.login"}, env=None)
browser.execute({"action": "type", "selector": "#user", "text": "alice"}, env=None)

Output sample

Navigated to https://example.com (200, 1.4s).

Lifecycle

The tool keeps a single browser context across calls in the same agent run. Closed automatically when the agent terminates. To force a fresh session, drop the tool from the agent’s tool list and re-instantiate.

See also

• read_image — for analyzing the screenshot output.
• web_fetch — text-only HTTP fetch.