Coasty – Computer-Use API at $0.05 per Prediction

TL;DR

TL;DR: Coasty is a computer-use API that drives any web or desktop application through natural-language instructions, operating the UI via screenshots and keyboard/mouse input — no API integration required.

Source and Accuracy Notes

⚠️ This section is MANDATORY. All links must be verified from actual source, not guessed.

• Project page: coasty.ai ← MUST visit and verify
• HN launch thread: news.ycombinator.com/item?id=48922706 ← MUST read
• Pricing source: coasty.ai/computer-use (JSON-LD schema verified 2026-08-01)
• License: Proprietary (no public OSS repo confirmed)

What Is Coasty?

Coasty is a computer-use AI API that lets developers send natural-language tasks and have an agent operate any web or desktop application through the visual interface — screenshots, mouse clicks, and keyboard input. The agent observes the current screen state, decides on the next action, executes it, then observes the result before continuing.

The core differentiator is breadth: the same API handles browsers, remote desktops, and legacy Windows applications without requiring DOM access, accessibility trees, selectors, or app-specific integrations.

From the HN launch post:

“A lot of important software is still difficult to automate. Healthcare teams submit prior authorizations through payer portals, accounting teams enter data into desktop applications, and operations teams move information between internal systems, spreadsheets, and remote desktops. Many of these applications have no API, incomplete APIs, or integrations that take months to build.”

Pricing (from the product page, verified via JSON-LD schema on 2026-08-01):

• Base price: $0.05 per /v1/predict call
• Published surcharges on top of base
• 1,000+ native app integrations
• Free sandbox keys
• OpenAPI 3.1 compliant
• $99/month Unlimited dashboard plan (flat rate, no per-seat)

How It Works

The Agent Loop

Coasty’s agent follows an observe-decide-act-verify loop:

1. Observe — receives a screenshot of the current screen state
2. Decide — selects the next UI action (click, type, scroll, etc.)
3. Act — executes the action via the target environment (browser, RDP, local app)
4. Verify — checks the resulting state before continuing to the next step

This is fundamentally different from traditional RPA, which records a fixed click sequence and replays it. RPA breaks when a button moves, a pop-up appears, or the application enters an unexpected state. The Coasty agent adapts dynamically.

API Interface

# Example: Submit a prior auth in a healthcare portal
curl -X POST https://api.coasty.ai/v1/predict \
  -H "Authorization: Bearer $COASTY_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "task": "Submit prior authorization for patient John Doe, procedure code 99213",
    "environment": "browser",
    "credentials": { "portal": "payer.example.com" },
    "screenshot_frequency": "after_each_action"
  }'

The response is a structured run record with:

• Screenshots at each step
• Actions taken (click coordinates, keystrokes, etc.)
• Outputs extracted
• Errors encountered

No DOM Required

Unlike browser automation tools that require selectors, XPath, or DOM access, Coasty works purely from visual output. This makes it applicable to:

• Legacy Windows applications (SAP, Citrix, Remote Desktop)
• Healthcare payer portals with no public API
• Accounting desktop software
• Any internal tool that lacks automation hooks

Practical Evaluation Checklist

• Can the API handle multi-step workflows (more than 5 steps)?
• Does it expose an OpenAPI 3.1 spec for easy integration?
• Are sandbox keys provided for testing without cost?
• Does the agent recover gracefully from unexpected UI states?
• Is the pricing model clearly documented (base + surcharges)?
• Are 1,000+ native app integrations verified and current?

Security Notes

• API key authentication required for all calls
• Credentials are stored and scoped per environment
• Screenshots and run records may contain sensitive PII — review data retention policies before using in healthcare or financial workflows
• The $99 Unlimited plan is a flat rate — confirm no hidden egress or storage charges for screenshot logs

FAQ

Q: How is this different from browser automation tools like Playwright or Selenium? A: Playwright and Selenium require the target application to have a structured DOM or API. Coasty operates purely from visual output (screenshots) and produces keyboard/mouse events. It can automate applications that have no web interface at all, including legacy Windows desktop software accessed via RDP.

Q: What happens if the agent encounters an unexpected popup or error dialog? A: The agent observes each action’s result before proceeding. If an unexpected state is detected, the agent can be configured to retry, escalate, or return an error rather than continuing blindly — unlike rigid RPA scripts that replay a fixed sequence.

Q: Is there a free tier? A: Free sandbox keys are available for testing. The $0.05 per predict call is the base API cost; surcharges are published on the pricing page. The $99/month Unlimited dashboard plan is a flat rate.

Q: Does Coasty have a GitHub repo? A: No. Coasty is a proprietary SaaS API. There is no public source repository. The codebase is not open source.

Conclusion

Coasty targets a specific gap in the automation landscape: applications with no API, no DOM, and no integration path — legacy healthcare portals, accounting desktop software, internal tools built on remote desktop protocols. At $0.05 per prediction with a flat $99 Unlimited dashboard plan and OpenAPI 3.1 compliance, it is positioned for developers who need to automate workflows that would otherwise require manual human intervention.

The 1,000+ native app integrations and free sandbox keys make it approachable for evaluation. If your automation problem involves an application you cannot modify or extend, Coasty is worth evaluating — particularly for healthcare, finance, and operations teams dealing with legacy systems.

Source and Accuracy Notes

• Project page: coasty.ai
• HN launch thread: news.ycombinator.com/item?id=48922706
• Pricing (verified via JSON-LD on product page, 2026-08-01): $0.05 /v1/predict base, published surcharges, 1,000+ integrations, free sandbox keys, OpenAPI 3.1, $99 Unlimited dashboard
• License: Proprietary (no public OSS repo)
• Source last checked: 2026-08-01