Coasty Computer Use API Pricing: Every Endpoint, Every Cent
You want an agent that watches the screen, reads UI text, and clicks buttons, just like a human. Traditional automation relies on brittle selectors or proprietary APIs, which break when layout changes. Coasty’s computer use API builds agents that see and act on the graphical interface directly. This post breaks down every endpoint, the exact cost per call, and how to meter your usage.
How it works
Coasty exposes a set of endpoints you call from your application. You send a screenshot, instruction, and a CUA version. The API returns actions like clicks, key presses, or text entry. You loop capture, predict, act until the status is done. For stateful runs and workflows, you start a session, then POST /v1/sessions/{id}/predict. You also have commands for grounding coordinates from element descriptions and parsing pyautogui code into structured actions.
curl https://coasty.ai/v1/predict \
-H "X-API-Key: $COASTY_API_KEY" \
-H "Content-Type: application/json" \
-d '{
"screenshot": "$(base64 -i screenshot.png)",
"instruction": "find the sign-up button and click it",
"cua_version": "v3"
}'Vision endpoints and costs
- ●POST /v1/predict costs $0.05 per call. Takes a base64 screenshot, an instruction, and cua_version. Returns actions and a status. Loop capture, predict, act until status is done.
- ●POST /v1/sessions costs $0.10 to create a session. Then POST /v1/sessions/{id}/predict costs $0.04 per call. This enables stateful trajectory memory across steps.
- ●POST /v1/ground costs $0.03. Takes a screenshot and an element description, maps it to x,y coordinates for precise targeting.
- ●POST /v1/parse is free. Turns pyautogui code into structured actions, useful for legacy automation scripts.
Vision calls are billed per API call, not per action.
Task runs and workflow costs
- ●POST /v1/runs launches a server-driven agent with machine_id, task, cua_version, instructions, system_prompt, max_steps, deadline_seconds, on_awaiting_human, and webhook_url. Each agent step costs $0.05.
- ●GET /v1/runs lists runs, GET /v1/runs/{id} returns details, POST /v1/runs/{id}/cancel cancels a run, POST /v1/runs/{id}/resume resumes it.
- ●GET /v1/runs/{id}/events streams Server-Sent Events with Last-Event-ID for reconnection. States include queued, running, awaiting_human, succeeded, failed, cancelled, timed_out.
- ●Workflows are a versioned JSON DSL of runs. POST /v1/workflows, POST /v1/workflows/{id}/runs, and POST /v1/workflows/runs create workflows and their runs. Task steps are billed $0.05 each. Step types: task, assert, if, loop, parallel, human_approval, retry, succeed, fail. Conditions and variables like {{inputs.x}} or stepId.field provide guardrails. Budget_cents, max_iterations, and deadline_seconds enforce hard limits.
Machines and billing
- ●POST /v1/machines provisions a cloud VM you can start, stop, and snapshot. The agent drives real desktops, browsers, and terminals, not just API calls.
- ●Billing is on a prepaid USD wallet where 1 credit equals $0.01. You pay per API call and per agent step based on the endpoints above.
- ●Webhooks are HMAC signed with header Coasty-Signature: t=unix,v1=hex. Use Idempotency-Key header for safe retries. Scopes gate key access.
- ●An MCP server lets you drive Coasty from Cursor, Claude Desktop, or other MCP clients.
Where this beats brittle automation
Traditional automation guesses CSS selectors or relies on undocumented APIs. A layout change breaks your script. Coasty’s computer use API reads the screen like a human does. The agent sees buttons, text, and layout, then decides where to click. This removes guesswork and reduces maintenance. You get a robust agent that adapts to UI changes without code updates.
You now have a clear view of every Coasty endpoint and its exact cost. Start building a computer use agent that sees, reasons, and acts on your desktop. Get a key at https://coasty.ai/developers and start automating with confidence.