Agent Catalog Generator
Enter agent metadata, get a catalog entry. No template, no snippet library, no code.
What the Generator Does
You fill in a form with the data you're already capturing anyway: agent type, owning team, model/provider, deployment tier, integrations, and the compliance disclosures from the governance certification. You also choose which teams the entry is for and what tone it should be written in.
What comes back is a complete catalog entry: headline, short tagline for the results list, capability summary, data scope, integrations, the compliance disclosures as flowing text, a recommended-teams note, and a short announcement snippet. All in English, ready to publish or edit further.
Nothing Is Invented
The generator works exclusively with what you entered. If a field is missing, it simply doesn't appear in the text - instead of making up a capability that isn't there.
Disclosures Stay Verbatim
The figures from the compliance certification and the chargeback rate are carried over unchanged, never rounded and never reworded.
Wording That Holds Up
The recommended-teams description covers what the agent's capabilities and data scope actually suit - not unsubstantiated claims about outcomes. That keeps the text free of overstated AI capability claims.
Three Steps to a Finished Text
- Enter agent data. A sample agent is pre-filled so you can see the flow right away.
- Choose the target teams. Target team, tone and length determine how the text sounds and how long it is.
- Generate. Within a few seconds the catalog entry is ready - with a button that copies it to the clipboard as Markdown.
If you need the same flow programmatically, find it in the Agent Catalog API and the guide From Agent Deployment to Catalog Entry.
Access: You need your API key to make the call. The form asks for it, uses it only for that one call, and never stores or logs it. To learn how to generate a key, see Authentication.
Why It Runs Through the Gateway
Between the form and the language model sits the Kong AI Gateway. It handles, in one place, what every application would otherwise have to solve on its own:
| Task | Where it's handled |
|---|---|
| Provider and model choice | at the gateway - switching providers changes nothing in the form |
| Access control | at the gateway, not in the application |
| Usage and cost | the gateway logs what was consumed, per call and per key |
| Protection against unwanted input | prompt guard at the gateway |
| Lowering response times | semantic caching for recurring requests |
The generator itself doesn't know a provider key. It knows an address.