Engineering

Coasty Computer Use API Pricing, Every Endpoint, Every Cent

Name: Coasty AI Employee
Brand: Coasty
Price: 19 USD
Availability: InStock
Rating: 4.8 (1250 reviews)

David Park|July 15, 2026|8 min

Ctrl+S

Most pricing docs list a flat fee and stop. You still have to guess how many steps a task will take. The Coasty computer use API is different. Every endpoint has a clear price per request or per step. You can estimate costs before you write a line of code. This guide covers every endpoint, every field, and the exact price you pay. We also show a working example with the real endpoint and environment variables.

How it works

The Coasty computer use API is stateless by default. You send a screenshot and an instruction. The model returns actions. You capture again and call predict again until the server returns a status of done. For stateful sessions, you create a session first. The session stores trajectory memory. You then call predict on that session ID. The server tracks state across calls. For element selection, ground maps a description to x,y coordinates. The parse endpoint turns pyautogui code into structured actions. Task runs let the server drive an agent to completion. Workflows let you compose runs into multi-step pipelines. Machines provision cloud VMs the agent can drive.

bash

# Example: Vision endpoint - $0.05 per call
# Read key from env, never hardcode it
export COASTY_API_KEY="your-key-here"

# Capture a screenshot (replace with your real path)
screenshot=$(base64 -i screenshot.png | tr -d '\n')

# Call /v1/predict
response=$(curl -s -X POST https://coasty.ai/v1/predict \
  -H "X-API-Key: $COASTY_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "screenshot": "'$screenshot'",
    "instruction": "Click the submit button",
    "cua_version": "v3"
  }')

# Print actions and status
echo "$response"

Vision endpoint

●POST /v1/predict
●Price: $0.05 per call
●Required fields: screenshot (base64 PNG), instruction (string), cua_version (string, default v3)
●Returns: actions (array of action objects), status (done or other)
●Loop: capture screenshot, call predict, act on actions until status is done

$0.05 per vision call. Capture, predict, act, repeat until status is done.

Session endpoint

●POST /v1/sessions - Create a stateful trajectory memory
●Price: $0.10 per session creation
●Required fields: none (optional machine_id, cua_version)
●Response: session_id (string)
●Then call POST /v1/sessions/{id}/predict ($0.04 per call)
●The session persists actions and state across predict calls

Ground endpoint

●POST /v1/ground
●Price: $0.03 per call
●Required fields: screenshot (base64 PNG), element_description (string)
●Returns: x, y coordinates (number) and element_id (string)
●Use to locate elements reliably without brittle selectors

Parse endpoint

●POST /v1/parse
●Price: Free
●Required fields: code (string, pyautogui syntax)
●Returns: structured actions (array of action objects)
●Convert legacy pyautogui scripts into Coasty actions

Task Runs endpoint

●POST /v1/runs - Start a server-side agent run
●Price: $0.05 per agent step
●Required fields: machine_id (string), task (string), cua_version (string, default v3)
●Optional: instructions (string, appended to base prompt), system_prompt (string), max_steps (integer), deadline_seconds (integer), on_awaiting_human (pause/fail/cancel), webhook_url (string)
●GET /v1/runs - List runs
●GET /v1/runs/{id} - Get run details
●POST /v1/runs/{id}/cancel - Cancel a run
●POST /v1/runs/{id}/resume - Resume a paused run
●GET /v1/runs/{id}/events - Stream Server-Sent Events with reconnect support via Last-Event-ID
●States: queued, running, awaiting_human, succeeded, failed, cancelled, timed_out
●Use v4 for autonomous mode with a pass/fail verifier

Workflows endpoint

●POST /v1/workflows - Create a versioned JSON DSL of runs
●POST /v1/workflows/{id}/runs - Execute a workflow
●POST /v1/workflows/runs - Ad-hoc inline workflow execution
●Step types: task, assert, if, loop, parallel, human_approval, retry, succeed, fail
●Conditions and variables use double-brace syntax (inputs.x, stepId.field)
●Hard guards: budget_cents, max_iterations, deadline_seconds
●Task steps are billed $0.05 each

Machines endpoint

●POST /v1/machines - Provision a cloud VM
●Optional: start, stop, snapshot operations
●The agent drives real desktops, browsers, and terminals, not just API calls

Billing and keys

●Prepaid USD wallet. 1 credit = $0.01.
●Auth header: X-API-Key: <key> or Authorization: Bearer <key>.
●Scopes gate keys.
●Idempotency-Key header makes writes safe to retry.
●Webhooks are HMAC signed with header Coasty-Signature: t=unix,v1=hex.
●MCP server lets you drive Coasty from Cursor, Claude Desktop, or other MCP clients.
●Common errors: 401 invalid key, 402 INSUFFICIENT_CREDITS, 403 INSUFFICIENT_SCOPE, 429 rate limit.

Where this beats brittle automation

Traditional automation relies on selectors that break when UI changes. The Coasty computer use API sees the screen and acts like a human. The model understands context, handles overlays, and adapts to layout shifts. You do not need to maintain fragile CSS selectors or APIs that do not exist yet. For tasks that involve clicks, forms, or browser interactions, computer use agents are more resilient. You pay only for steps you actually use, with transparent pricing per endpoint.

You now know every Coasty computer use API endpoint, every field, and the exact price. Build reliable UI agents, workflows, and browser tasks with clear cost expectations. Get a key at https://coasty.ai/developers and start running your first task run.