Engineering

Coasty Computer Use API Pricing: Every Endpoint and Every Cent

Name: Coasty AI Employee
Brand: Coasty
Rating: 4.8 (1250 reviews)

Emily Watson|August 3, 2026|6 min

⌘+W

You do not want to guess the bill for an agent that watches screens and clicks buttons. The Coasty computer use API charges by the action, not by the hour. Every endpoint has a clear price in cents. This guide lists every endpoint, the fields you send, the fields you get back, and the exact cost. You can estimate your bill before you write a single line of code.

Pricing overview

●Wallet model: 1 credit = 1 USD cent. Prepaid wallet, no subscriptions.
●Vision call: POST /v1/predict costs $0.05 per request.
●Stateful session: POST /v1/sessions costs $0.10, then POST /v1/sessions/{id}/predict costs $0.04 per step.
●Grounding: POST /v1/ground costs $0.03 per request.
●Parse endpoint: POST /v1/parse is free.
●Task run (server-driven agent): POST /v1/runs bills $0.05 per agent step.
●Workflow DSL: task steps bill $0.05 each.
●Machines: POST /v1/machines provisions a cloud VM, billed separately by the provider.

Vision endpoint: /v1/predict

Use this when you want to drive a session yourself. You send a base64 screenshot, an instruction, and a cua_version. The API returns actions and a status. You can loop capture, predict, and act until status is "done".

bash

curl -X POST https://coasty.ai/v1/predict \
  -H "X-API-Key: $COASTY_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "screenshot": "$(base64 -i screenshot.png | tr -d "\n")",
    "instruction": "click the ok button",
    "cua_version": "v3"
  }'

Sized session endpoint: /v1/sessions

Create a session that stores trajectory memory. You get an id. Then call POST /v1/sessions/{id}/predict for each step. The session remembers what happened before, so the next call can reason about the full history.

Session step endpoint: /v1/sessions/{id}/predict

●Cost $0.04 per session step.
●You still send a screenshot and an instruction, but you do not need cua_version in this call.
●The response includes actions and status, just like the vision endpoint.

Vision endpoint: $0.05. Session creation: $0.10. Session step: $0.04.

Grounding endpoint: /v1/ground

●Maps a screenshot plus an element description to x,y coordinates.
●Useful when you want to target a UI element by text instead of brittle selectors.
●Cost $0.03 per request.

python

import base64
import requests
import os

url = "https://coasty.ai/v1/ground"
key = os.environ["COASTY_API_KEY"]

with open("screen.png", "rb") as f:
    base64_img = base64.b64encode(f.read()).decode()

payload = {
    "screenshot": base64_img,
    "description": "submit button"
}

response = requests.post(url, headers={
    "X-API-Key": key,
    "Content-Type": "application/json"
}, json=payload)

print(response.json())

Parse endpoint: /v1/parse

●Accepts raw pyautogui code and returns structured actions.
●Use this when you want to refactor legacy scripts.
●This endpoint is free.

Task runs: /v1/runs

Let the server drive an agent to completion. You send machine_id, task, cua_version, and optional instructions or system_prompt. You can set max_steps, deadline_seconds, and what to do when the agent waits for human approval. The API bills $0.05 per agent step. You can stream events with GET /v1/runs/{id}/events.

bash

curl -X POST https://coasty.ai/v1/runs \
  -H "X-API-Key: $COASTY_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "machine_id": "m-abc123",
    "task": "open chrome and navigate to https://coasty.ai",
    "cua_version": "v4",
    "max_steps": 10,
    "deadline_seconds": 300,
    "on_awaiting_human": "pause"
  }'

Workflows DSL

●Define a versioned JSON workflow with tasks, asserts, conditionals, loops, and parallel steps.
●Task steps are billed $0.05 each.
●Use variables like {{inputs.x}} and stepId.field for dynamic values.
●You can POST /v1/workflows, /v1/workflows/{id}/runs, or inline runs with POST /v1/workflows/runs.

Where this beats brittle automation

APIs change. IDs shift. Fixed selectors break when a UI refreshes. A computer use agent watches the screen, maps elements by description, and clicks like a human. You still pay per action, but you do not need to maintain fragile selectors. Grounding gives you precise coordinates from natural language. The parse endpoint lets you turn existing scripts into structured actions. Workflows let you chain tasks with retries, conditions, and human approval.

You now know every endpoint and every cent for the Coasty computer use API. Build agents that see and act, estimate your bill, and scale with confidence. Get your key at https://coasty.ai/developers.