← 返回命令列表

Linux command

rotunda 命令

文本

复制后可按需替换文件名、目录或参数。

常用示例

Download

uvx rotunda agent fetch

Example

uvx rotunda agent new-profile --name [agent-demo]

Example

uvx rotunda agent new-context [agent-demo]

Navigate

uvx rotunda agent navigate [3] [https://example.com]

Describe

uvx rotunda agent describe [3]

Click

uvx rotunda agent click [3] [ref_id]

Example

uvx rotunda agent screenshot [3] --path [page.png]

说明

Rotunda is a browser purpose-built for AI agents. It bundles a stealthy Chromium build, exposes high-level primitives for navigation, inspection, and form interaction, and persists profiles, daemon sessions, and short resource indexes under ~/.rotunda so that later uvx rotunda calls can attach to the same profile without rebooting it. The agent CLI is grouped into a few families of subcommands: profile and session management (fetch, new-profile, new-context, stop), navigation primitives (navigate, back, forward, reload, wait), inspection (describe, screenshot, extract), and interaction (click, hover, fill, type, select, check, press, upload, dialog). Each element-targeted command refers to a numeric context id and an element reference returned by describe. Rotunda is also importable as a Python library on top of Playwright, but the CLI is the recommended entry point for shell-based agents.

参数

fetch
Download the bundled browser build to ~/.rotunda.
new-profile --name _name_
Create a persistent browser profile.
new-context _profile_
Open a fresh browser context in the named profile; prints a numeric context id.
navigate _ctx_ _url_
Load _url_ in the given context.
describe _ctx_
Print page structure and reference ids that other commands consume.
click _ctx_ _ref_, hover _ctx_ _ref_
Mouse interaction with a referenced element.
fill _ctx_ _ref_ _text_
Replace the value of an input field.
type _ctx_ _text_
Type text at the current focus position.
select _ctx_ _ref_ _value_
Choose a dropdown option by value.
check _ctx_ _ref_, press _ctx_ _key_
Toggle a checkbox / send a keyboard key.
extract _ctx_ _--format_ _text|html|markdown|links_
Pull text, raw HTML, markdown, or link lists out of the page.
screenshot _ctx_ _--full-page_ _--path_ _file_
Capture the viewport or the entire page.
upload _ctx_ _ref_ _file_
Attach _file_ to a file-input element.
dialog _ctx_ _action_
Accept or dismiss browser dialogs (alert/confirm/prompt).
wait _ctx_ _condition_
Block until a page condition is met (load state, selector, timeout).
back, forward, reload _ctx_
Standard navigation primitives.
stop
Shut down the profile daemon.

FAQ

What is the rotunda command used for?

Rotunda is a browser purpose-built for AI agents. It bundles a stealthy Chromium build, exposes high-level primitives for navigation, inspection, and form interaction, and persists profiles, daemon sessions, and short resource indexes under ~/.rotunda so that later uvx rotunda calls can attach to the same profile without rebooting it. The agent CLI is grouped into a few families of subcommands: profile and session management (fetch, new-profile, new-context, stop), navigation primitives (navigate, back, forward, reload, wait), inspection (describe, screenshot, extract), and interaction (click, hover, fill, type, select, check, press, upload, dialog). Each element-targeted command refers to a numeric context id and an element reference returned by describe. Rotunda is also importable as a Python library on top of Playwright, but the CLI is the recommended entry point for shell-based agents.

How do I run a basic rotunda example?

Run `uvx rotunda agent fetch` in a terminal, then adjust file names, paths, flags, or remote targets for your system.

What does fetch do in rotunda?

Download the bundled browser build to ~/.rotunda.