Hi HN, I built a draft protocol called AIP (Agentic Internet Protocol).
The premise: instead of trying to make agents use human websites, build agent-only websites from the ground up.
In AIP:
- each “page” is a small plain-text node (no HTML/CSS)
- each node includes Title, Description, Content, and a small Edges menu (recommended ~6)
- sites are navigated as a graph (nodes + edges)
- edges are typed:
- NAV = go to a known node
- QRY = generate a node from parameters (search/filter/pagination)
- ACT = change state (cart/checkout/payment)
- pages stay small (well under ~8k words) to reduce cost/latency and make prompt injection harder to hide
Repo includes:
- the draft spec
- two hero sites (ecommerce + billing) as AIP nodes
- a tiny Python parser + interactive “node browser” demo
I’m looking for feedback on:
- the minimal node format (too strict? not strict enough?)
- edge typing (NAV/QRY/ACT) and whether it matches how you’d build agent surfaces
- what’s missing for real-world adoption (auth, pagination conventions, tooling)
tylersuard•1h ago
The premise: instead of trying to make agents use human websites, build agent-only websites from the ground up.
In AIP:
- each “page” is a small plain-text node (no HTML/CSS) - each node includes Title, Description, Content, and a small Edges menu (recommended ~6) - sites are navigated as a graph (nodes + edges) - edges are typed:
Repo includes:- the draft spec - two hero sites (ecommerce + billing) as AIP nodes - a tiny Python parser + interactive “node browser” demo
I’m looking for feedback on:
- the minimal node format (too strict? not strict enough?) - edge typing (NAV/QRY/ACT) and whether it matches how you’d build agent surfaces - what’s missing for real-world adoption (auth, pagination conventions, tooling)