Skip to main content

AI Services Catalog

Model routers pick
the model.
We pick everything else.

Search, scraping, enrichment, filings, translation, data. One endpoint, one bill, pay per call.

See the catalog in the app.

Intelligence relies on data

  • Liveness probed hourly
  • Rich endpoint schemas
  • Semantic indexing
  • Rank Intelligence
  • Auto mode
Station 1 · a need goes in

Your agent asks for

latest 10-K for NVIDIA, plain text

one need, one call. example

An illustration. 23 provider marks from the catalog are drawn as a faint band, standing in for 162 live listings. A line passes across them, every mark fades, and one is left at full strength inside a ring: EDGAR (SEC), at its listed price of $0.008, charged $0.00816 with the 2% fee on top. This is an example of one need being resolved to one service, not a live result.

You never saw the shelf. That is the point.

Bought direct, you choose the provider yourself, sign up, and find out which ones answer after you have paid. Here the need went in and one service came back, called and settled.

No model reads the listings and no model chooses. The pick is arithmetic, and you get the score it won on.

Auto is live today. One need resolves to one service per call, at the listed price with 2% on top.

Three ways to ask

  • Name it

    pay_service

    You already know the service. Call it and pay.

  • Price it

    quote_service

    What would this cost. Charges nothing, signs nothing, reserves nothing.

  • Hand it over

    route_by_intent

    Say the need. This is what happened above.

Station 2 · the shelf

You do not browse this shelf. You ask it.

One brief, six steps. Each need goes in as a plain sentence and comes back as one listing at its listed price. No account per service, no key per service, no shortlist to keep.

Three of the six show what they were chosen out of. Names and prices are live from the catalog. The run itself is illustrative.

  1. ask 01

    Find the funding news for every company on my list.

    returned

    Exa Search

    $0.007 / request

    also scored, not returned

    • Linkup API

    • Serper

    • Tavily API

    Cheapest did not win and was never going to. Every listing in the catalog is indexed on meaning and on words, and one ranker serves every surface, so what comes back answers the sentence rather than the price list.

  2. ask 02

    Give me the clean text of each page that turns up.

    returned

    Firecrawl Scrape

    $0.002 / scrape

  3. ask 03

    Some of those pages only load in a real browser.

    returned

    Browserbase

    $0.01

    also scored, not returned

    • Browser Use

    • Steel

    • Notte

  4. ask 04

    Capture each pricing page as an image I can paste.

    returned

    ScreenshotOne

    $0.055

  5. ask 05

    Tell me who is hiring and who just shipped something.

    returned

    PredictLeads

    $0.04

    also scored, not returned

    • Coresignal

    • People Data Labs

    • Apollo API

  6. ask 06

    Read the latest annual filing for the public ones.

    returned

    EDGAR (SEC)

    $0.008

The other three asks were scored across the same catalog. They are shown resolved, without their candidates.

6 services, at their listed prices
$0.1220
Our fee, 2%
$0.0024
The brief, one charge
$0.1244

One charge, whichever services answered. The 9 candidates that did not come back are not on it, and there is nothing up front and no minimum.

The same brief, fifty companies

$6.22

Fifty companies before lunch, for the price of a coffee you did not buy. That is the number a procurement conversation for six vendors never gets to.

The width of the catalog only matters because it is what the winning answer was chosen out of.

162 scored, 6 returned.

Breadth is the denominator, not the offer. The candidates above are the shelf: real listings, scored on the same brief, that did not come back and cost nothing.

Browse the catalog
  • Data & Enrichment35
  • AI & Media18
  • Web Scraping & Automation18
  • Search & Research16
  • Sales & Business Intelligence14
  • Finance & Markets8
  • Social & Creator8
  • Infrastructure & Compute7
  • Communication & Voice5
  • Identity & Compliance2
  • Travel1
  • Weather1

The catalog is the shelf we stock and vouch for. Listed or not, the same call pays any paid endpoint on the public internet, so the same key reaches past the end of it.

Station 3 · what you own

what a shelf costs you

9 services is not 9 purchases.

4 chores follow every one of them: account, key, card on file, invoice. One row per chore, one column per service. Count them.

9 services bought direct, each with its own account, key, card on file, invoice:

  • Exa Search
  • Firecrawl
  • Browserbase
  • ScreenshotOne
  • Serper
  • PredictLeads
  • People Data Labs
  • EDGAR (SEC)
  • DeepL

That is 4 chores each and 36 things to own. Through the catalog it is 3: one account, one card and one invoice, whether it is 9 services or ninety.

Not one account, one card and one invoice for each of them. One of each, for all of them, and a ceiling you set.

Station 4 · the ceiling

“my agent runs up a bill while I sleep”

Set the ceiling. Then go to sleep.

You are the only one who can move it. Go on, try it from the other side.

monthly ceiling

example · changes no real account

$25.00

a budget is three terms: $25.00 · expires in 30 days · 200 purchases. no per transaction limit.

you moved it

0

your agent moved it

0

you

$5.00$500.00

One click revokes it, and nothing further can be charged.

your agent

not asked yet

press it as often as you like

The answer does not change. The only way the ceiling goes up is a person at a browser, with a card, raising it themselves. That is not a missing feature. That is the product.

Every call is 2% on top of the service price. Nothing else, and nothing taken from the service.

Station 5 · one purchase

One purchase, and every call draws on it.

You have counted twice on this page already. This is the third count, and the last: buy one pack and all 162 services on the shelf draw on it, whichever one answers the call.

Bought direct

36

accounts, keys, cards and invoices to own

Then, through the router

3

one account, one endpoint, one invoice

Then, with credits

1

one purchase, and nothing to settle at month end

One pack

bought once, for $10

10,000

credits, illustrative

An illustrative draw sheet: one pack of 10,000 credits, and the credits left after each of 8 calls.
the calldrawncredits left
a web search89,992
a page scraped129,980
a company filing looked up49,976
a page opened in a hosted browser319,945
a screenshot taken39,942
a company record enriched179,925
a paragraph translated29,923
one more search, on what came back89,915
and so on, for the rest of the pack

8 calls in, and 85 credits of 10,000 are gone. The dim digits are the ones that never changed.

A cap is money. A pack is a count of calls.

Budget

$25.00

Currency, on your card. The cap you set in the station above, which your agent cannot raise.

no total

Pack

10,000credits

A count, not an amount. What the pack holds, drawn down one call at a time.

There is no number that is both units, so nothing on this page adds them together. The cap stays where you set it, in currency, on the card. Credits is the count behind it: a pack of 10,000 credits costs $10, which is roughly a thousand calls at the price most catalog services charge.

You pay by card exactly as you would for anything else. Credits are the unit the calls are counted in, and every service on the shelf counts in the same one, so a pack bought for a search job covers a scrape, a filing lookup or a translation just as well.

When a pack runs low you re-order it, or let it renew on its own. Figures above are an example, not a quote.

the end of the wire

One line, and your agent can buy everything else.

Everything above ends at one address. Paste it in, and your agent has the whole shelf.

Make your first callIt starts in the chat you already have open.

Twelve tools arrive the moment you paste it. Ten search, compare and price a job without moving a cent. Two spend, and only under a ceiling you set.

a hosted assistant

Paste it into your connectors and approve it once, in your own browser. ChatGPT, Claude and Claude Code today.

a coding agent

claude mcp add --transport http nevermined <endpoint>, and the same one line in Cursor.

your own code

npm i @nevermined-io/payments, or call the REST API directly.

A hosted assistant gets nothing at all: you approve it yourself, in your own browser, and only hosts we have listed can be approved in the first place. Anything else reaches the Approve button and is turned away.

Selling a service instead? List it once and every agent on this endpoint can find it, price it and pay for it. Talk to us about listing.