Bulk corpus instead of a sweep
Training crawlers get one corpus export and a delta feed. The class most likely to change behaviour for a large efficiency gain is exactly the class hammering you today.
For mid-market SaaS, content sites and publishers
See every machine visitor. Serve each one correctly.Take the crawls off your origin.
You have rich dynamic content and crawlers that hit it hard. The result is unpredictable compute spikes, database bottlenecks during indexing sweeps, and no way to tell which agent took what. ADN classifies that traffic at the edge, serves each class the shape its job needs, and answers most of it without your origin being involved at all.
- Full nine-class taxonomy
- Bulk export + delta feed
- Content Signals
- Web Bot Auth
Sound familiar?
Where it hurts today
- 01
Compute spikes you did not schedule
A training crawler walking your sitemap page by page renders every one of them. On serverless that is a bill; on a database it is a connection pool.
- 02
Every bot gets the same answer
One Markdown format for all machine visitors means a training crawler and a live fetch with a human waiting are treated identically, and neither is served well.
- 03
No idea who took what
Bot analytics stop at 'automated traffic'. Which agent, which class, how many pages, what they were served — most site owners have never seen it.
What you get
What ADN does about it
Per-class policy
Rate limits, terms, metering and response shape set per intent class, not per user agent string — with legacy SEO crawlers passed through untouched.
Proof against your own baseline
Traffic-by-class analytics, origin requests avoided, and compute saved measured against what you served before offload went live.
Fail-open by construction
An edge connector is about twenty lines in your edge and every failure path is a plain fetch to your origin. Humans are never on our critical path, and you keep the kill switch.
Two ways in
Whatever is in front of your origin
The pitch changes with your infrastructure, not the product. On Cloudflare we install alongside your zone plan; everywhere else we are the whole agent layer.
Path 1 — Cloudflare add-on
Already on Cloudflare Pro
Keep Pro. Add intent routing, bulk corpus and origin offload as a Snippet installed over OAuth — your WAF, DDoS protection and caching stay exactly as they are.
Pro $25 + Growth $149 ≈ $174/mo
Path 2 — primary stack
On Vercel, Netlify or AWS
No Cloudflare plan underneath, so ADN is the entire agent layer — npm middleware with a restrictive matcher, priced below the add-on path.
Growth — $149/mo
Where you land
Growth$149/mo
Hosted proxy or edge connector
3 domains, 750k requests, 5k renders, policy engine, bulk corpus. Scale ($449, coming soon) adds multi-hostname hosted proxy, commercial metering, and an SLA.
If you need an SLA today
Scale — multi-hostname hosted proxy, commercial metering, instant purge, and a signed SLA — is coming soon. An SLA is a commitment we will not make before we can keep it. If that is your requirement now, the honest path is a design partnership, not a purchase order. And if the actual problem is credential stuffing or fraud, that is Bot Management: a different product, bought by a different team.
Free efficiency scan
Get the numbers for your own domain
The scan estimates the origin requests your sitemap generates under repeated crawls, and what those pages cost in tokens as raw HTML.
Questions
The ones that come up first
- Won't letting bots through cost me more?
- It is the right question, and the arithmetic is short. For a crawler you block today, Cloudflare returns a 403 and your origin does nothing; afterwards an edge connector forwards it to us and your origin still does nothing — roughly no change. For a crawler you allow today, which is most bots on most sites, your origin renders the page at full compute and egress cost; afterwards it is never touched. That gap is the ROI.
- What happens when we hit our quota?
- The gateway fails open and traffic passes through to your origin. You keep 100% availability and see the load return — an upgrade prompt rather than an outage.
- We deploy middleware on Vercel. Any billing surprises?
- One, and we would rather you hear it from us: middleware invocations bill as Edge Requests. Install with a restrictive matcher so the edge connector only runs on the paths that matter — a loose matcher inflates your Vercel bill and it will look like our fault.
- Is this Bot Management?
- No, and we will not pitch it as a cheaper one. Bot Management scores how bot-like a request is and feeds WAF rules for fraud. ADN classifies what a machine is trying to accomplish and routes content accordingly. Sites frequently want both.
Get the numbers for your own domain
The scan estimates the origin requests your sitemap generates under repeated crawls, and what those pages cost in tokens as raw HTML.
Not quite you?