Turn your highest-intent conversations
into revenue.

Built for conversations. Not retrofitted for them.

Under 300ms at p95, and never in front of your model call

Decision latency is under 300ms at p95. The request is dispatched in parallel with your completion, not ahead of it, so your response streams on its own schedule and the slot fills when the decision returns. If it does not return in time, the slot stays empty and the conversation is unchanged. An ad server outage is invisible to your users by design, and you can verify that by pointing the SDK at a dead endpoint.

From integration to revenue.

Your rendering context decides which demand can serve

Impression-priced demand requires a real browser or app context where the impression can be verified client-side. Headless surfaces monetize on outcomes instead. We publish this because it determines what you earn.

React, iOS, and Android SDKs are in development. An impression fired from a server originates in datacenter IP space and is classified as invalid traffic, which is why headless surfaces are priced on outcomes rather than impressions. Full detail at choosing your integration path.

<script src="https://api.adgentek.ai/storage/v1/object/public/sdk/sdk.js"></script>
<script>
  AdgentekAds.init({ apiKey: 'YOUR_FULL_API_KEY_HERE', slotId: 'YOUR_SLOT_ID' });
</script>

<div id="adgentek-ad"></div>

// call after each AI response
AdgentekAds.requestAd([
  { role: 'user', content: userMessage },
  { role: 'assistant', content: aiResponse }
]);

Your API key is generated in the dashboard under Settings, API Keys and is prefixed adgt_. The slot ID is the placement ID from Inventory, Surfaces, and it drives targeting, floor pricing, and reporting. Placeholder credentials do not serve ads.

{
  "mcpServers": {
    "adsmcp": {
      "url": "https://mcp.adgentek.ai/mcp",
      "headers": { "Authorization": "Bearer YOUR_PUBLISHER_TOKEN" }
    }
  }
}

Transport is MCP Streamable HTTP. The serve tool registers as ads.serve and clients sanitize it to ads_serve. Listed in the official MCP Registry as io.github.Adgentek/adsmcp.

Complete documentation is at docs.adgentek.ai. Coding agents should start from the Adgentek integration skill and use the docs index rather than a cached snippet.

Ad types built for agentic AI. Not borrowed from web pages.

Every format is designed to feel native to conversational AI interfaces. Users experience them as helpful, not intrusive.

Operators who have already built both sides of this stack

Adgentek's founders have been building programmatic infrastructure since 2007, one on the sell side and one on the buy side.

Yield management, waterfall logic, floor pricing, discrepancy handling, and the demand relationships that make a fill rate real are not new problems. They are the problems this team spent two decades solving before AI surfaces existed. Adgentek is that stack rebuilt for a conversational surface, not a first attempt at ad serving.

Four tiers of demand, not one sales team

A single direct-sold demand pool leaves most requests unfilled. Adgentek runs a hard waterfall, so a request that does not clear the direct tier keeps falling until it finds a buyer rather than returning empty.

Estimate your revenue

Your assumptions, our math. Adjust every field to match your traffic.

Privacy-safe. Semantic. Real-time.

No cookies. No user profiles. No PII. Just real-time understanding of what the user is actually talking about.

Your app. Your rules.

We're infrastructure. You decide how monetization works in your product.

Works across every AI surface.

Guides for your surface

Each surface has a different latency budget, session shape, and set of formats that fit. These go deeper than this page does.

Frequently asked questions

Built to protect your users and your brand.