Working MVP, October 2026

AI agents should pay for the journalism they use. We’re building the market where they do.

Clear Exchange is a private exchange for licensed content. An agent sends a question and a price ceiling. We route it to the sources most likely to answer, and their articles come back ranked by relevance and priced from the publishers’ own rate cards. The agent buys a signed entitlement to one article, retrieves it, and tells us how it used it. We record the sale, our fee and the publisher’s share.

Run a query in the demo Or read the example below

A worked example

One question, from query to settlement

This is real output from the exchange, run on 28 October 2026 against its seeded catalogue. The publishers in that catalogue are fictional; the API, ranking, prices and signatures are the ones that run in production. Long IDs are shortened.

The agent asks

A research assistant wants a short excerpt it can show a user. It will pay up to a cent per article, our fee included. Anything dearer is removed from the list, not pushed down it.

Request POST /v1/query

{
  "query": "Do heat pumps still work in very cold climates?",
  "agent_id": "research-assistant-01",
  "max_price": { "amount": "0.010", "currency": "USD" },
  "usage": "display_snippet",
  "limit": 5
}

Publishers answer

Sixteen candidate articles. Eight were too far off the question and dropped, however cheap. Two relevant ones cost $0.0132, over the ceiling, so they were excluded. The response says so, and gives that price, so the agent can decide whether to pay more.

Relevance sets the order. Price competes only among articles nearly as relevant as the best one: within 90% of its score, and at least 0.50. The cheapest of those is selected. Here the threshold was 0.56, only the Fieldnote piece cleared it, so it is the selection.

Each offer carries its licence terms, so the agent knows what it may do before it buys.

Offers returned, ranked by relevance (BM25, 0 to 1). Prices are all-in: publisher price plus our fee.
#	Article	Relevance	Price	Licence terms
1	Fieldnote ScienceDo heat pumps work in cold climates? What the field data shows Selected	0.62	$0.0055	Citation required; excerpt up to 400 characters; keep 7 days; no AI training
2	Cobble & ByteSmart thermostats and heat pumps: getting the settings right	0.48	$0.0033	Attribution, no citation needed; excerpt up to 500 characters; keep 30 days; no AI training
3	The Harbour LedgerHow regional banks are financing the heat pump installation boom	0.39	$0.0044	Citation required; excerpt up to 320 characters; keep 24 hours; no AI training. Archive discount applied
4–5	Two weaker matches from Fieldnote Science	0.33, 0.31	$0.0055	As row 1
Not shown: 8 below minimum relevance (0.20); 2 relevant but over the ceiling, cheapest at $0.0132 with relevance 0.40. Band threshold 0.56 (90% of 0.62); one offer in the band.

The agent buys one article

Accepting the selected offer returns an entitlement: a token signed with Ed25519 that names the agent, the article, the usage, the price and how it splits between publisher and exchange, the terms and a 15-minute window. Anyone can check it offline against our published keys; the publisher does not have to trust our database.

Entitlement, decoded POST /v1/offers/off_309f…/accept

// header
{ "alg": "EdDSA", "typ": "cx-entitlement+jwt", "kid": "cx-FtMBpU0W…" }
// claims (iss, aud, iat, nbf and hashes left out)
{
  "sub": "research-assistant-01",
  "jti": "ent_13938e8e…",
  "exp": 1793186220,
  "cx": {
    "v": 1,
    "content_id": "fn-2026-0153",
    "publisher_id": "pub_fieldnote",
    "usage": "display_snippet",
    "terms": {
      "citation_required": true,
      "attribution": "Fieldnote Science",
      "max_snippet_chars": 400,
      "retention_seconds": 604800,
      "ai_training": "prohibited"
    },
    "price": { "amount_micros": 5500, "currency": "USD" },
    "clearing": {
      "publisher_price_micros": 5000,
      "fee_micros": 500,
      "fee_bps": 1000
    },
    "rate_card_ref": "rc_fieldnote@v1"
  }
}

Retrieval, reporting, settlement

The article is served only against a valid entitlement. Retrieval writes a ledger line. The agent then reports what it did with the text as SPUR Content Telemetry events: used it to ground an answer, cited it, showed it to a user, and whether the user clicked through.

Ledger line for this retrieval. The clearing fee is 10% on top of the publisher’s price, a placeholder while we test pricing.
Publisher	Usage	Gross	Fee	Publisher net
Fieldnote Science	Display snippet	$0.005500	$0.000500	$0.005000

content_retrieved by the exchange
content_grounded by the agent
content_cited by the agent
content_presented by the agent
content_engaged by the agent

For publishers

Your price, your terms, and a record of every use

You set the price: A rate card per usage tier (grounding only, a display snippet, full text), with rules by section, tag or article age. Archive pieces can cost less than this morning’s story.
Relevance first, then price: The best answer to the question comes first. Price competes only among articles nearly as relevant as the best one, and the cheapest of those is selected. A cheaper but weaker article does not jump the queue. Our fee goes on top of your price; it is not taken out of it.
Terms travel with the article: Citation, attribution, excerpt length, how long the agent may keep the text, and whether it may train on it are written into every entitlement.
You see what happened next: Whether your article was used to ground an answer, cited, shown to a person, or clicked, reported per retrieval in SPUR Content Telemetry.

See the demo publishers’ rate cards

For AI companies and agent builders

Licensed sources through one API, paid per retrieval

One integration: Query once. We send the question to the few connected sources most likely to answer, including open content from Wikinews and Wikipedia alongside paid publishers, and wait at most 1.5 seconds for them.
A budget you control: Set an all-in ceiling per query and the usage you need. You pay only for what you retrieve, and a priced-out query tells you the price that would have cleared.
Terms a machine can follow: Every offer states its terms as data, so your agent can respect citation, excerpt and retention limits without a lawyer in the loop.
Proof you were licensed: Signed entitlements (EdDSA JWS) you can keep, verifiable against our public key set.

Try the agent console

What runs today

Live, and on the roadmap

We would rather be precise than impressive. This is what the service does now and what we are working on next.

Capabilities and their status
Capability	Status	Detail
Query, ranking and offers	Live	BM25 relevance sets the order; the cheapest offer within 90% of the best match is selected. Over-budget offers are excluded.
Query routing	Live	Each query goes to the sources most likely to answer, chosen from a thin profile of each (not their content), and is fanned out live within a 1.5-second budget.
Publisher offer API	Live	A publisher that implements our v1 offer contract is connected by configuration (`http-offer`). Five fictional demo publishers use it today.
Signed entitlements	Live	Ed25519 JWS, keys at /.well-known/jwks.json.
SPUR Content Telemetry	Live	SPUR Coalition Content Telemetry v1.0 events, authorised by the entitlement and deduplicated.
Ledger and clearing	Live	Gross, fee and publisher net per retrieval. The fee is 10% on top of the publisher price, a placeholder. No money moves yet.
Confidential-VM attestation	Roadmap	Run the exchange in a confidential VM so publishers and buyers can check, by remote attestation, which code handles their content.
Prebid and Trusted Server seller module	Roadmap	An open-source module so publishers can offer content from infrastructure they already run (IAB Tech Lab Trusted Server, Prebid).
Agent identity	Roadmap	Web Bot Auth signatures in place of self-declared agent IDs.

If you publish, or build agents that read, I’d like to talk.

We are lining up the first publishers and buyers for live supply. A call is the quickest way to see whether the pricing and terms work for you. The demo is open to anyone and runs on the real API.

Paul, founder

Book a call Open the demo