Traditional automation relies on brittle selectors and rigid APIs. You might wait for a DOM element to exist, but if a layout shifts or a class changes, the script breaks. The /v1/predict endpoint solves this by giving you a computer use agent that sees the screen and acts like a human. You send a base64 screenshot and an instruction, and it returns actions such as click, double-click, scroll, and type. You loop capture, predict, act until the status is done. This is how you build a computer use agent that works across any application, website, or desktop tool.
How /v1/predict works
The endpoint is a single HTTP POST request. You authenticate with an API key in the X-API-Key header. The payload includes a base64-encoded screenshot, an instruction string, and the cua_version to specify the model. The response contains the actions array, the status field, and a request_id for debugging. You run the actions with your own automation library, then capture a new screenshot and call predict again. You repeat until status is done. This loop gives you a stateful trajectory memory without managing sessions yourself.
# Example using curl (replace with your own base64 screenshot and API key)
export COASTY_API_KEY="your_api_key_here"
BASE64_SCREENSHOT="$(base64 -i screenshot.png | tr -d '\n')"
curl -X POST https://coasty.ai/v1/predict \
-H "X-API-Key: $COASTY_API_KEY" \
-H "Content-Type: application/json" \
-d '{
"screenshot": "'"$BASE64_SCREENSHOT"'",
"instruction": "Click the Apply button on the top right and enter my email.",
"cua_version": "v3"
}' | jq .Request and response fields
- screenshot (string): base64-encoded image. Required.
- instruction (string): natural language command for the agent. Required.
- cua_version (string): model version. Use v3 for the default agent.
- actions (array of objects): each object contains action type, x coordinate, y coordinate, text, and optional keys. Returned on success.
- status (string): either done or an error code. Loop until done.
- request_id (string): unique request identifier for debugging.
- price (number): the cost in credits for this prediction. It is $0.05 per call.
Loop capture, predict, act until status is done.
Where /v1/predict beats brittle automation
A computer use agent reads the screen like a person. It can click a button that changes its label, scroll through a long list, or type into a field that has no unique ID. It does not depend on a specific CSS selector or API endpoint. If the UI updates, the agent still sees the change and adapts its actions. This makes the /v1/predict endpoint ideal for automating legacy systems, complex web forms, or tools that lack a public API. You get a computer use agent that works across browsers, desktop apps, and terminals without brittle selectors.
The /v1/predict endpoint gives you a simple, powerful way to turn screenshots into actions. Build a computer use agent that clicks, types, and scrolls across any application. Sign up for a key at https://coasty.ai/developers and start automating with computer use.
Want to see this in action?
View Case Studies