agent‑ads.org is a feed and manifest layer for agent‑native advertising. You publish a signed manifest; AI agents discover it, check its provenance, and rank it in an open broadcast feed — no scraping, no guesswork.
abf-index.json ·
open viewer →
Three moving parts, in plain terms. The protocol names are kept in parentheses — you don't need them to understand the system.
A small JSON file at /.well-known/agent-ad.json that declares who you are and what you're sponsoring — signed, so it can be attributed back to you (the covenant manifest).
Before anything is trusted, its source and signature are checked against an open schema. Sponsorship lineage is mandatory and traceable (Trust Lineage / CALT).
Verified signals are published as a machine‑readable feed agents poll for eligibility and ranking — available as JSON and RSS (the Agent Broadcast Feed, ABF).
What this isn't: it's not an ad network that places ads for you, and it doesn't decide what your agent does. The feed is data an agent can read — never an instruction it must follow.
One path, four hops. Each hop is inspectable — you can see exactly what was published, validated, and ranked.
You host a signed agent-ad.json on your own domain.
It's checked against the open schema, and its signature and source are verified.
Valid entries are merged into the broadcast feed with provenance attached.
Agents poll the feed and rank entries by relevance and freshness.
This is the architectural commitment that separates agent‑native sponsorship from traditional ad‑tech — and the principle the rest of the system is built to enforce. It isn't a slogan; it's a placement model.
Make your sponsorship legible to AI systems without buying placements.
Declare what's sponsored on your pages so agents can disclose it downstream.
Read the schema, the disclosure minimums, and the feed spec — all open.
Poll a lightweight index, fetch the canonical feed, rank what's relevant.
Trust isn't asserted — it's constructed from five checkable properties. If any fails, the entry doesn't broadcast.
Every entry carries a watermark tracing it to its origin domain and task.
Sponsorship must be declared — the disclosure minimums spell out the baseline fields.
Manifests validate against a public JSON Schema before they're accepted.
Each entry is timestamped; agents can weight or drop stale signals.
The operator is named, not hidden — part of the Velastra constellation. Appearing in the feed is a signal, not an endorsement.
Everything humans read above maps to a concrete endpoint. These are real, pollable, and documented.
The covenant manifest you publish — identity + sponsorship declaration.
Canonical Agent Broadcast Feed — full items with provenance and trust.
Lightweight index for cheap hourly polling — fetch the full feed only on change.
Human‑ and reader‑friendly rankings as an RSS feed.
The public JSON Schema you validate your manifest against.
Plain‑text orientation for agents arriving at the domain.
Go live with a compliant /.well-known/agent-ad.json in an afternoon.
Copy a minimal manifest and adapt the fields to your domain and sponsorship.
Run it through the validator. When $schema is present, detection is automatic.
Serve it at the well‑known path, then point robots, sitemap, and headers at it so agents can find it.
These terms appear across the agent‑facing layer. You can build against the system without memorizing any of them.
Inspect real entries, their provenance, and their raw JSON — the same data an agent retrieves when it polls the broadcast feed.