What Exa built
One new endpoint. That’s the whole server-side delta.| Field | Value |
|---|---|
| Endpoint | POST https://admin-api.exa.ai/team-management/nevermined/purchase-key |
| Auth | payment-signature header carrying an x402 access token |
| Price | $7 per purchase, charged via Nevermined card delegation (nvm:card-delegation scheme) |
| Response | { "status": "ok", "apiKey": "...", "expiresAt": null } |
| Top-up | Same endpoint, same payer wallet → existing key tops up instead of issuing a new one |
| When exhausted | Regular Exa endpoints return HTTP 402 + { "tag": "NO_MORE_CREDITS" } |
| Endpoint type | Calls per $7 |
|---|---|
| Standard search | ~1,000 |
| Deep-Lite search | ~700 |
| Content pages | ~7,000 |
| Deep-Reasoning | ~466 |
Why this is strategically interesting for Nevermined
Most of what Nevermined had been doing before this is monetizing agents — paywalling the AI service itself, charging per call. Exa flips the angle: now Nevermined is monetizing what agents pay for to do their job. That’s a much wider surface. Every agent that runs end-to-end consumes paid services along the way — search, document retrieval, image generation, market data feeds, maps, weather, financial APIs. Each of those is a potential Exa-shaped integration. And because the provider-side cost is so low (one endpoint, no metering changes), the pattern travels well.“We’re no longer just monetizing the agents themselves — we’re monetizing paid services that agents may need to complete their task. And best of all the integration is really simple.”
How agents discover the integration
Exa’s llms.txt lists their Nevermined integration under## Integrations — a single line linking to a static nevermined.md. That’s the entire discovery surface. Any agent that reads the llms.txt (and most LLM-powered tools that fetch URLs do) follows the link, reads the integration page, and now knows how to pay for and use Exa.
The integration page itself is six sections: intro / buy a key / what $7 buys / sample request / parameters from Nevermined / when the key runs out / references. Nothing more. The structure is so small that any LLM-class agent can ingest it as part of normal context and act on it without special prompting.
How to publish your own llms.txt + nevermined.md →
What this means for other providers
If you’re an API provider with existing metering, the takeaways:Integration cost is genuinely small
One purchase endpoint, no per-call settlement, no SDK on your hot path. The handler is the same shape regardless of which card rail (Stripe / Braintree / Visa Trusted Agent Protocol) the buyer enrolled.
You keep your pricing surface
Exa kept their internal metering, their rate limits, their tier structure. Nevermined just adds a new top-up trigger. You can offer the same product to humans and to agents with no fork in the codebase.
Agent discovery is static
No agent registry to maintain, no API to call. A
llms.txt + a nevermined.md on your existing docs site is the whole thing.The pattern is reusable
Exa’s integration page mirrors a structure you can adopt verbatim — same section headings, same fields. Agents that have seen one will navigate yours.
Try it yourself
Theweb-search-paid-api-ts tutorial implements the Exa pattern end-to-end against a stub web-search service. Server, client, llms.txt + nevermined.md templates, README walkthrough. Clone it, point it at a sandbox PAYG plan, and you’ll have an agent autonomously buying an API key in under five minutes.
Related
- Exa’s live integration page
- Exa’s
llms.txt - Monetize paid APIs for agents — the pattern this case study ships
- Agent discoverability — how to publish your own integration page
- API Providers — implementation guide — the technical deep-dive