Coasty Computer Use API Pricing: Every Endpoint, Every Cent
You want a computer use agent that drives real desktops, browsers, and terminals. You need to know exactly how much each request costs so you can forecast bills and tune workloads. Coasty exposes a set of endpoints for vision, grounding, stateful sessions, and server-driven task runs, each with its own price. This post lists every endpoint, the exact request fields, and the real cents-per-call cost.
Base URL and Authentication
All requests go to https://coasty.ai/v1. Use an X-API-Key header or Authorization: Bearer <key>. Read the key from COASTY_API_KEY environment variable. Keys live at https://coasty.ai/developers/keys.
curl https://coasty.ai/v1/predict \
-H "X-API-Key: $COASTY_API_KEY" \
-H "Content-Type: application/json" \
-d '{
"screenshot": "data:image/png;base64...",
"instruction": "Click the Apply button.",
"cua_version": "v3"
}'Vision API: predict endpoint
- ●Endpoint: POST /v1/predict
- ●Cost: $0.05 per call
- ●Request fields (all required):
- ● screenshot (base64 PNG string)
- ● instruction (what to do)
- ● cua_version (string, e.g., 'v3')
- ●Response fields:
- ● actions (array of structured actions)
- ● status (string, loop until 'done')
Grounding API: map screenshot + description to coordinates
- ●Endpoint: POST /v1/ground
- ●Cost: $0.03 per call
- ●Request fields:
- ● screenshot (base64 PNG string)
- ● element_description (what to locate)
- ●Response fields:
- ● x (integer, screen coordinates)
- ● y (integer, screen coordinates)
Stateful trajectory memory: sessions
- ●Endpoint: POST /v1/sessions
- ●Cost: $0.10 for the initial session creation
- ●Response fields:
- ● id (string, session identifier)
- ●Subsequent: POST /v1/sessions/{id}/predict
- ●Cost: $0.04 per predict call
- ●Request fields:
- ● instruction (what to do)
- ● cua_version (string, e.g., 'v3')
- ●Response fields:
- ● actions (array of structured actions)
- ● status (string, loop until 'done')
Task Runs: server-driven agent
- ●Endpoint: POST /v1/runs
- ●Cost: $0.05 per agent step
- ●Request fields:
- ● machine_id (string, VM identifier)
- ● task (string, what the agent does)
- ● cua_version (string, 'v3' or 'v4' for autonomous with pass/fail verifier)
- ● instructions (string, appended to base prompt)
- ● system_prompt (optional string)
- ● max_steps (integer, max API calls)
- ● deadline_seconds (integer, timeout)
- ● on_awaiting_human ('pause', 'fail', or 'cancel')
- ● webhook_url (optional)
- ●Response fields:
- ● id (string, run identifier)
- ● state (string: queued, running, awaiting_human, succeeded, failed, cancelled, timed_out)
- ●Subsequent: GET /v1/runs/{id} for status
- ●Stream: GET /v1/runs/{id}/events for SSE events, reconnect with Last-Event-ID
Workflows: versioned JSON DSL
- ●Endpoint: POST /v1/workflows, POST /v1/workflows/{id}/runs, POST /v1/workflows/runs
- ●Cost: $0.05 per task step inside a workflow
- ●Step types: task, assert, if, loop, parallel, human_approval, retry, succeed, fail
- ●Variables: {{inputs.x}} and stepId.field
- ●Hard guards: budget_cents, max_iterations, deadline_seconds
Machines: provision cloud VMs
- ●Endpoint: POST /v1/machines
- ●Cost: Provision cost applies per VM (not listed as per-call, see docs)
- ●Response fields:
- ● id (string, machine identifier)
- ● status (string: provisioning, running, stopped, etc.)
- ●Operations: start, stop, snapshot
Parse endpoint: pyautogui to structured actions
- ●Endpoint: POST /v1/parse
- ●Cost: Free
- ●Request fields:
- ● pyautogui_script (string, Python pyautogui code)
- ●Response fields:
- ● actions (array of structured actions)
Billing and safety
- ●Prepaid USD wallet: 1 credit = $0.01
- ●Webhooks: HMAC signed header Coasty-Signature: t=unix,v1=hex
- ●Idempotency-Key header for safe retries
- ●Scopes gate key permissions
- ●MCP server lets you drive Coasty from Cursor, Claude Desktop, and other MCP clients
Errors
- ●401: invalid key
- ●402: INSUFFICIENT_CREDITS
- ●403: INSUFFICIENT_SCOPE
- ●429: rate limit
- ●Always JSON with error.code, error.message, error.request_id
Every Coasty endpoint has a clear, published price: $0.03, $0.04, $0.05, $0.10. Start building a computer use agent with predictable billing at https://coasty.ai/developers.
Why computer use beats brittle automation
Traditional automation relies on brittle selectors and API-only tools. Computer use agents see the screen, interpret instructions, and act like a human. They navigate dynamic UIs, handle unexpected layouts, and adapt to changes without code updates. This makes them ideal for tasks that span apps, websites, and terminals.
Now you know every endpoint and its cents. Build a computer use agent that drives real desktops, browsers, and terminals with predictable billing. Get a key at https://coasty.ai/developers. Start coding.