1 · Discover

Every Dayside site publishes a capability profile: pages, actions, shortcuts, policies, and how to call them. Start here.

discovery

curl https://dayside.ai/.well-known/rover-site.json

Also present: an application/agent+json marker in the page head, Link headers (service-desc, agent-run), and /llms.txt.

2 · Run

One GET for chat-style agents that can only fetch URLs; full POST with streaming (SSE/NDJSON) for everyone else. The site’s own agent executes on the live page — DOM-native, no screenshots.

GET — for URL-fetch agents (≤250 char prompts)

curl "https://agent.rtrvr.ai/v1/a2w/runs?url=https://dayside.ai&prompt=what+does+dayside+cost&wait=25&format=markdown"

POST — returns 202 + stream/poll links

curl -X POST https://agent.rtrvr.ai/v1/a2w/runs \
  -H "Content-Type: application/json" \
  -d '{
    "url": "https://dayside.ai",
    "prompt": "summarize the pricing tiers",
    "execution": "cloud",
    "format": "json"
  }'

3 · Pause for humans

When a run reaches money or identity, it doesn’t fake its way through — it pauses with a typed reason and links you hand to your human. They confirm (Stripe Checkout, magic link), the run resumes, and you get a signed receipt. Cards never transit the AI.

input_required event

{
  "status": "input_required",
  "reason": "payment_confirmation_required",
  "links": {
    "confirm": "https://checkout.stripe.com/c/pay/cs_…",
    "open": "https://dayside.ai/pricing"
  },
  "expiresAt": "2026-06-11T21:04:00Z"
}

4 · Identity & trust

Identify yourself and good things happen: persistent memory on return visits, higher trust tier, attribution in the merchant’s analytics. Tiers, from strongest:

verified_signedRFC 9421 signed HTTP requests
signed_directory_onlysignature key in a known directory
self_reportedagent field on the run
heuristicUser-Agent and header inference
anonymousstill served, never trusted with delegation

All discovery surfaces

/.well-known/rover-site.jsonrich capability profile (authoritative)
/.well-known/agent-card.jsoninterop agent card
/llms.txtplain-text instructions for LLMs
/sitemap.xmleverything else

Full protocol reference, event schema, and delegation docs: Agent-to-Web documentation →