Publisher Intelligence

Your content powers AI.
You should get something back.

Relay sits between your site and AI agents — detecting crawlers, serving structured content at 94% lower bandwidth cost, and giving you full visibility into who's reading what.

Free up to 100K events/month. No credit card required.

94%
bandwidth savings per AI request
38K
pages per referral visit by AI agents
<5ms
added latency at the edge
20+
AI agents detected and classified

AI agents read everything.
Publishers pay the bill.

AI companies train and run models on publisher content. They fetch thousands of pages per session, consume full HTML — scripts, ads, nav — and send zero revenue back.

Without Relay

Full HTML to every crawler

A single GPT training run fetches 38,000 pages. Each page is ~180KB of HTML. You pay for every byte — bandwidth, CDN egress, origin CPU — and receive nothing in return.

180KB × 38,000 pages = 6.8 GB per session
At $0.09/GB CDN egress → $0.61 per AI visit
With Relay

Structured content, fraction of the cost

Relay intercepts agent requests at the edge and serves a compact Structured Object Model (SOM) — pure content JSON, no scripts, no ads, no nav. Agents get exactly what they need.

180KB HTML → ~11KB SOM JSON
94% bandwidth reduction per request

Deploy once. Works everywhere.

Relay runs as a Cloudflare Worker — no code changes, no server restarts. Install in under 5 minutes with one command.

01

Install the Worker

Run npx relay-init in your terminal. The CLI handles wrangler setup, API key, KV namespace, and deployment automatically.

02

Agents are detected

The Worker identifies AI agents by user-agent pattern, behavioral signals, and request patterns. 20+ agents classified — GPT, Claude, Gemini, Perplexity, and more.

03

SOM served, data collected

Agents receive structured JSON instead of full HTML. Your dashboard shows every request — which agent, which page, what tier, how many bytes saved — in real time.

What agents receive

Instead of parsing megabytes of HTML, agents get a precise content graph. Faster for them. Cheaper for you. Richer for everyone.

Full HTML response ~180KB
<!DOCTYPE html> <html lang="en"> <head> <meta charset="UTF-8"/> <script src="/analytics.js"></script> <script src="/ads.bundle.js"></script> <link rel="stylesheet" href="/main.css"/> <title>How photosynthesis works</title> </head> <body> <nav class="navbar"> <div class="nav-logo">...</div> <ul class="nav-menu"> <li><a href="/">Home</a></li> <li><a href="/science">Science</a></li> </ul> </nav> <div class="ad-banner" id="gpt-slot-1">...</div> <article> <h1>How photosynthesis works</h1> <p>Photosynthesis is the process...</p> </article> </body></html>
Relay SOM response ~11KB
{ "relay": { "version": "1.0", "format_served": "som", "cost_saved_bytes": 171842, "license_policy": "attribution-required" }, "document": { "url": "https://example.com/science/photosynthesis", "title": "How photosynthesis works", "description": "Complete guide to photosynthesis...", "author": "Dr. Sarah Chen", "published": "2024-03-15", "content_type": "article" }, "body": [ { "type": "heading", "level": 1, "text": "How photosynthesis works" }, { "type": "paragraph", "text": "Photosynthesis is the process by which plants use sunlight, water and CO₂ to produce oxygen and energy..." } ] }

Built for every kind of publisher

Whether you run a news site, a SaaS docs portal, or an API platform, Relay gives you control over how AI accesses your content.

Media Publishers

News sites, blogs, and content platforms bearing the full bandwidth cost of AI training crawls — with zero revenue from it.

  • See every training crawler by name
  • Enforce attribution or contact licensing
  • Cut CDN egress costs by 94%

SaaS Documentation

Developer docs and knowledge bases that want AI assistants to accurately represent their product — not outdated crawled summaries.

  • Serve structured docs to coding agents
  • Control which agents can index what
  • Analytics on which docs AI reads most

API & Data Providers

Platforms whose structured data is the source material for AI outputs — and who need a programmatic, monetizable access layer.

  • License data to AI companies directly
  • Block unauthorized scraping by policy
  • Full audit log of agent access

Start free. Scale as you grow.

Every plan includes agent detection, SOM serving, and full analytics. Upgrade when your traffic does.

Free
$0/mo
Up to 100K events/month
  • Agent detection
  • SOM serving
  • Analytics dashboard
  • 1 domain
Scale
$999/mo
Up to 25M events/month
  • Everything in Growth
  • Advanced analytics exports
  • Unlimited domains
  • Dedicated Slack channel
Enterprise
Custom
Unlimited events
  • Everything in Scale
  • Custom licensing contracts
  • On-prem / private deployment
  • SLA + dedicated support

Common questions

Does Relay require code changes to my site?+
No. Relay runs entirely as a Cloudflare Worker. You run npx relay-init once, add a route in your Cloudflare dashboard, and it's live. Your origin server is untouched.
Does it affect my human visitors?+
Not at all. Relay only intercepts requests it identifies as AI agents. Human traffic passes straight through with zero modification and sub-millisecond overhead.
What is SOM and how is it generated?+
SOM (Structured Object Model) is a compact JSON representation of your page's content. The Worker generates it in real time by extracting JSON-LD, Open Graph metadata, and semantic HTML — stripping scripts, ads, and navigation. Result is typically 94% smaller than full HTML.
Can I block specific AI agents entirely?+
Yes. Set your license policy to blocked in the Settings tab. The Worker returns a 403 with a structured license JSON explaining the policy. You can also set contact to redirect agents to your licensing page.
Does Relay work without Cloudflare?+
Currently Relay deploys as a Cloudflare Worker — your domain needs to be on Cloudflare (free plan works). Support for additional edge runtimes (Fastly, AWS Lambda@Edge) is on the roadmap.
Is my content stored by Relay?+
No. Relay only stores traffic metadata — which agent, which URL, timestamp, bytes saved. The content of your pages is never written to Relay's database. SOM responses are cached briefly in Cloudflare KV (5 minutes) and then discarded.
What counts as an "event"?+
One event = one AI agent request to your domain that Relay intercepts. Human traffic is never counted. Each request — whether it gets a SOM response, a passthrough, or a block — counts as one event.

AI agents are crawling your site right now.

Find out which ones, what they're reading, and how much it's costing you — free, in under 5 minutes.