Browser and desktop

Uni-CLI groups browser and desktop actions under stable command families. Start with a state read, select a concrete target, then perform the action.

Browser sessions

Check the local setup.

unicli browser doctor --json

Start a hidden managed browser.

unicli browser start

Use an existing Chrome session in the background.

unicli browser --background start

The mode applies to the current CLI invocation. Include --background on each follow-up command to keep using the existing Chrome provider. The following unqualified commands use the default hidden managed browser.

Open a page and inspect its accessible state.

unicli browser open https://example.com
unicli browser state -f json

The state response assigns refs to interactive elements. Pass a ref to actions such as click, type, and query.

unicli browser click <ref>
unicli browser type <ref> "search text"

Desktop applications

Search by the result you want.

unicli search "inspect a desktop application" --surface desktop

Read compute health before the first desktop session.

unicli doctor compute --json

Take a snapshot, then act on a returned ref.

unicli compute snapshot --app "Calculator" -f json
unicli compute click <ref> --app "Calculator" -f json

Available providers depend on the operating system and installed accessibility services. The health response lists the active provider and the setup action for each unavailable provider.

Native coordinate driver

Cua Driver provides the explicit --via driver path for pixel-based desktop actions. Run the provider probe to install a missing driver through its official installer and report macOS Accessibility and Screen Recording permissions.

unicli doctor compute --providers --install --json
unicli compute screenshot --via driver -f json

Inspect the screenshot, choose the target in its pixel coordinates, and pass .data.observation.ref to compute point-click <x> <y> --observation <ref> --via driver. Each observation expires after 30 seconds and permits one coordinate action. Capture a new screenshot after the visible target changes or the driver restarts.

Uni-CLI forwards the capture identity, driver session, and screenshot coordinates to Cua Driver. The driver applies Retina scaling and the coordinate transform for size-capped captures. Use coordinates from the returned image without applying an additional display-scale adjustment. Missing permissions and provider failures return structured recovery information with the next setup action.

Choose the smallest useful interface

Structured APIs and native CLIs are efficient for data and repeatable actions. Browser semantics are useful for signed-in pages. Desktop accessibility reaches installed applications. Visual actions cover interfaces that expose only pixels. unicli search reports the operator selected by each catalog entry.