You want an agent that can click, type, and scroll like a human, not a brittle selector or API call. The Coasty Computer Use API gives you two core model versions, v3 and v4, through the same endpoint. v3 is a stateless loop where you call /v1/predict with a screenshot and instruction. v4 wraps that loop into a single request with an autonomous pass/fail verifier. Pick v3 for tight control and v4 when you want a guaranteed outcome with less code.
How it works
Both v3 and v4 use the POST /v1/runs endpoint to launch a task. The cua_version field decides the model behavior. v3 is stateless, you provide a machine_id, task, and optionally instructions. The server runs the agent and streams events until success, failure, or cancellation. v4 adds a pass/fail verifier and runs autonomously until it reaches a terminal state. The API bills $0.05 per agent step for both versions. The response includes status, events, and a run_id you can use to poll or cancel later.
curl -X POST https://coasty.ai/v1/runs \ -H "X-API-Key: $COASTY_API_KEY" \ -H "Content-Type: application/json" \ -d '{
"machine_id": "vm-123",
"task": "Open Chrome, navigate to https://coasty.ai, and click the Developers link",
"cua_version": "v4",
"max_steps": 100,
"deadline_seconds": 300,
"on_awaiting_human": "pause"
}'v3 vs v4 at a glance
- Both versions use the same POST /v1/runs endpoint.
- cua_version field switches between v3 and v4.
- v3 is stateless, you control the loop and action history.
- v4 runs autonomously with an included pass/fail verifier.
- Both bill $0.05 per agent step.
- v4 reduces code for guaranteed completion scenarios.
Use v3 for fine-grained control and v4 when you need an autonomous pass/fail verifier.
Where this beats brittle automation
Traditional automation relies on selectors, IDs, and stable URLs. When a UI changes layout, the selector breaks and the script fails. A computer use agent sees the screen, understands context, and adapts to layout shifts. The API sends screenshots and actions, so the agent can click visually correct elements even if they move. This visual, human-like behavior works across browsers, terminals, and apps without brittle selectors.
Pick v3 when you want to step through actions and debug, or v4 when you want an autonomous agent that guarantees success. Start building with the Computer Use API today. Get your key at https://coasty.ai/developers.
Want to see this in action?
View Case Studies