Coasty AI Computer Use Agent API
Coasty is an AI agent API for computer-use automation. Score 85.60% on OSWorld benchmark, 1000+ app integrations, and $0.05 per predict call. YC S26.
TL;DR
TL;DR: Coasty is a YC S26-backed computer-use agent API that controls desktop and web applications via screenshots, mouse, and keyboard — no DOM access or selectors needed.
Source and Accuracy Notes
This section is MANDATORY. All links must be verified from actual source, not guessed.
- Project page: coasty.ai — verified July 16, 2026
- API docs: coasty.ai/docs — verified July 16, 2026
- OpenAPI spec: coasty.ai/openapi.json — verified July 16, 2026
- Pricing: coasty.ai/pricing — verified July 16, 2026
- HN launch thread: news.ycombinator.com/item?id=48922706
- License: Proprietary (YC S26 startup, no open-source license confirmed)
What Is Coasty?
Coasty is a computer-use AI agent platform that operates applications the way a human does — through screenshots, mouse clicks, and keyboard input. Unlike traditional RPA tools that require pre-recorded sequences or DOM selectors, Coasty observes the current screen state, decides on an action, executes it, then verifies the result before continuing.
According to the product description on the website:
“AI computer-use agent that controls a desktop like a human (85.60% on OSWorld) and ships the best computer-use API of 2026 at $0.05 per /v1/predict, with 1,000+ native app integrations (Gmail, Slack, Salesforce, HubSpot, Linear, GitHub, Stripe), free sandbox keys, OpenAPI 3.1 spec, and a $99 flat Unlimited dashboard plan.”
The agent handles legacy desktop software, web applications without usable APIs, and any interface where traditional automation breaks down. The OpenAPI 3.1 spec is publicly available at coasty.ai/openapi.json.
API Overview
The Coasty Public API (v1.0.0) exposes endpoints for:
/v1/predict— Send a natural-language task and receive action predictions with screenshots/v1/triggers/webhook/{webhook_id}— Webhook-triggered automation runs (no auth required)- Managed VM provisioning for isolated execution environments
- Scheduled automation against provisioned environments
Authentication is via API key passed in request headers. The /v1/triggers/webhook/{webhook_id} endpoint and health checks are the only exceptions.
Setup Workflow
Step 1: Get an API Key
Sign up at coasty.ai to receive free sandbox API keys with initial credits.
Step 2: Make Your First Request
curl -X POST https://coasty.ai/v1/predict \
-H "Authorization: Bearer YOUR_API_KEY" \
-H "Content-Type: application/json" \
-d '{
"task": "Submit a prior authorization for patient John Doe in the payer portal",
"environment": "browser",
"credentials": {
"portal_url": "https://payer.example.com"
}
}'
Step 3: Integrate with MCP
Coasty provides an MCP server for integration with Claude, Cursor, and other MCP-compatible clients. Configure your MCP client to point at https://api.coasty.ai/mcp.
Pricing
Confirmed from the product page schema on July 16, 2026:
| Plan | Price | Credits/month | |------|-------|---------------| | Starter | $19/mo | 200 | | Plus | $50/mo | 600 | | Unlimited | $99/mo | Unlimited |
Per-call pricing: $0.05 per /v1/predict call. This is the primary metered usage model, with monthly plans providing bundled credits.
Practical Evaluation Checklist
- [ ] OpenAPI spec is publicly accessible and validates
- [ ] API key authentication works as documented
- [ ]
/v1/predictaccepts natural-language task descriptions - [ ] Managed VM provisioning is functional
- [ ] Webhook trigger endpoint responds without auth
- [ ] OSWorld benchmark claim (85.60%) is verified against official benchmarks
- [ ] 1,000+ integrations list is current and accessible
- [ ] Sandbox keys are provided at no cost for testing
Security Notes
- API keys grant access to execute actions inside managed VMs — treat them like database credentials
- Webhook endpoint (
/v1/triggers/webhook/{webhook_id}) requires no authentication — use unique webhook IDs per integration to prevent guessing - Credentials passed in requests are stored in Coasty’s managed environment during the run duration
- Review Coasty’s data retention policy before sending sensitive business data through the API
FAQ
Q: What is OSWorld? A: OSWorld is a benchmark for evaluating AI agents’ ability to complete real-world tasks on desktop operating systems. A score of 85.60% places Coasty among the top computer-use agents as of 2026.
Q: How is Coasty different from traditional RPA? A: RPA tools like UiPath or Automation Anywhere require pre-recorded sequences or DOM selectors specific to each application. Coasty uses a vision-based agent that can generalize to new interfaces without application-specific integrations.
Q: Does Coasty require DOM access or browser extensions? A: No. Coasty operates purely through screen observation, mouse, and keyboard — no DOM access, accessibility tree, selectors, or browser extensions are required.
Q: Can Coasty handle slow-loading pages or unexpected pop-ups? A: Yes. The agent observes the resulting state after each action before continuing, so it handles dynamic UI elements, loading states, and pop-ups that break fixed-sequence RPA.
Q: Is there an open-source version? A: No. Coasty is a proprietary commercial product (YC S26). No open-source license was confirmed at time of publication.
Conclusion
Coasty fills the gap for automating legacy software and web applications without APIs — the category where traditional RPA breaks down but intelligent agents can still operate. At $0.05 per predict call with an OpenAPI 3.1 spec and MCP server support, it is designed to fit into developer workflows rather than requiring a separate RPA platform.
If you have an automation scenario where the application cannot be modified, no API exists, and a fixed sequence of steps is insufficient, Coasty is worth evaluating. Start with the free sandbox credits to test against your actual use case before committing to a paid plan.
Related Posts
dev-tools
Automotive Skills Suite for AI Engineering
Evaluate Automotive Skills Suite for APQP, ASPICE, HARA, safety-plan, and DIA workflows with setup notes, governance risks, and SME review guidance.
5/28/2026
dev-tools
awesome-agentic-ai-zh Roadmap Guide
Explore awesome-agentic-ai-zh as a Chinese agentic AI learning roadmap, with setup notes, track selection, study workflow, and evaluation guidance.
5/28/2026
dev-tools
Baguette iOS Simulator Automation Guide
Set up Baguette for iOS Simulator automation, web dashboards, device farms, gesture input, streaming, and camera testing with Xcode caveats.
5/28/2026