You want an agent that can click, type, and navigate real apps, not just mock selectors. The Coasty Computer Use API lets you pick a cua_version when you start a task run. v3 is a guided agent. v4 is autonomous with a pass/fail verifier. Pick the right model to balance cost, control, and reliability.
How it works
When you launch a task run you POST to /v1/runs. The cua_version field tells the server which agent model to drive the session. v3 means a guided session where you can add instructions and wait for human approval. v4 is autonomous with a verifier that decides if the task succeeded. The API returns a run_id, status, and events stream. You poll /v1/runs/{id} or stream /v1/runs/{id}/events to see progress. Billing is $0.05 per agent step.
#!/bin/bash
# Launch a v4 autonomous task run
COASTY_API_KEY="${COASTY_API_KEY}"
BASE_URL="https://coasty.ai/v1"
TASK_URL="${BASE_URL}/runs"
# Example task: open a browser and navigate to a URL
TASK='{
"machine_id": "${MACHINE_ID}",
"task": "Open Chrome, navigate to https://coasty.ai, and wait for the page to load.",
"cua_version": "v4",
"instructions": "Ensure the page title contains Coasty.",
"max_steps": 100,
"deadline_seconds": 300
}
run_id=$(curl -s -X POST "${TASK_URL}" \
-H "Authorization: Bearer ${COASTY_API_KEY}" \
-H "Content-Type: application/json" \
-d "${TASK}" | jq -r .run_id)
echo "Started run: ${run_id}"
# Poll the run status
while true; do
status=$(curl -s -X GET "${BASE_URL}/runs/${run_id}" \
-H "Authorization: Bearer ${COASTY_API_KEY}" | jq -r .status)
echo "Status: ${status}"
if [[ "${status}" == "succeeded" || "${status}" == "failed" || "${status}" == "cancelled" ]]; then
break
fi
sleep 5
done
echo "Run finished with status: ${status}"What cua_version does for you
- v3 is a guided session. You can append instructions to the base prompt, use on_awaiting_human to pause or fail when the agent needs help, and control the flow with max_steps and deadline_seconds.
- v4 is autonomous with a built-in verifier. The agent decides when the task is done and reports pass or fail. This reduces manual checks and is ideal for repeatable, supervised tasks.
- Both versions bill at $0.05 per agent step, but v4 removes the need for you to poll and inspect intermediate states, saving development time and operational overhead.
The cua_version field determines guided vs autonomous behavior in a task run.
Where this beats brittle automation
Traditional automation relies on brittle selectors, hardcoded IDs, and API-only tools. A change in a site layout or UI version breaks those selectors. The Coasty Computer Use API lets agents see the screen and act like a human. They understand visual context, handle dynamic layouts, and can adapt to UI changes without you rewriting selectors. v4 goes further by automatically verifying success, so you can ship production agents with confidence.
Pick v3 for guided workflows where you want control and human oversight. Pick v4 for autonomous tasks where the agent drives the full workflow and self-verifies success. Start building with the Coasty Computer Use API at https://coasty.ai/developers to get your API key.
Want to see this in action?
View Case Studies