Licensed data · AI inference · One API
The prism is the same: raw results still come apart into clean, typed bands. What leaves the other side is no longer only JSON — it is analysis, grounded in licensed data and computed on hardware we own.
The intelligence layer
Raw results are a commodity. What is not a commodity is licensed data with analysis computed on top of it — same envelope, same key, same bill.
Being built in this release. The endpoints, request shapes and credit prices below are already fixed in code, which is why we can publish them — but every card is marked soon because none of them is serving traffic yet. What is live today is further down the page: the six data bands, the twelve inference tools and catalog extraction.
Family 1 of 3
Grounding that arrives ready to inject: deduplicated, reranked, cited, and cut to your token budget.
POST /v1/context
One call turns a query into grounding blocks that already fit your context window — deduplicated, reranked and citable.
curl -X POST https://api.dataswap.io/v1/context \
-d '{ "query": "vector database pricing", "token_budget": 4000 }'POST /v1/extract/schema
Give it a JSON schema and it comes back filled from the live web, with a citation and a confidence per field — null where there is no evidence.
curl -X POST https://api.dataswap.io/v1/extract/schema \
-d '{ "query": "acme robotics funding", "json_schema": { … } }'npx @dataswap.io/mcp
One command wires any MCP agent to licensed search plus the analysis tools — no glue code, no scraper in your supply chain.
npx @dataswap.io/mcp
# tools exposed: search · context · entity · schema_fill · answer_landscapeFamily 2 of 3
The answer layer is the new results page. These two measure who the engines actually cite, and what they say about your shelf.
POST /v1/geo/answer-landscape
One call returns the organic top-10, the AI Overview and AI Mode with their citations, and sampled LLM answers — with share-of-answer computed in code, not asked of a model.
curl -X POST https://api.dataswap.io/v1/geo/answer-landscape \
-d '{ "query": "best home ev charger", "brand": "example.com" }'POST /v1/commerce/shelf
Share of digital shelf for a whole category in one snapshot: who is in the shopping units and at what price, who ranks, and which products the AI Overview names.
curl -X POST https://api.dataswap.io/v1/commerce/shelf \
-d '{ "keywords": ["trail running shoes"], "market": "pt" }'Family 3 of 3
Everything about one entity, captured at a single timestamp — and a receipt that says where it came from.
POST /v1/entity
A domain-360 in one call: search presence, backlinks, technologies, reviews and registration, all read at the same instant instead of stitched from three vendors on three days.
curl -X POST https://api.dataswap.io/v1/entity \
-d '{ "domain": "example.com" }'POST /v1/commerce/match
The most expensive question in e-commerce as an API: is this offer the same product as that one? Verdict same / variant / bundle, with a confidence and the evidence.
curl -X POST https://api.dataswap.io/v1/commerce/match \
-d '{ "product": { "title": "Trail Runner GTX 42" } }'POST /v1/provenance/verify
Every response can carry a signed manifest of where it came from, and anyone can verify it without an API key — including the legal team that has to approve you.
# Public and keyless — whoever audits you has no key.
curl -X POST https://api.dataswap.io/v1/provenance/verify \
-d '{ "receipt": { … }, "payload": { … } }'The Authorization: Bearer sk_… header is omitted above for width, not because it is optional. Composed calls quote a ceiling and settle on the real cost, so you are never billed for parts that did not run — see pricing.
Answer Landscape
For one query we read the organic results, the AI Overview and AI Mode with their citations, and sampled answers from three LLMs — then compute share of answer per domain in code.
“best home ev charger”Sample data, not a live call — this page never touches the API. AI Overview and AI Mode are read from the results page and are deterministic; LLM answers are not, so they are sampled over 12 prompts and the methodology travels with the response.
Who cites techradar.com
How much the engines agree on the set they cite. Low consensus means the answer is still up for grabs.
Live today
The eight capabilities above are new. This part is not: these twelve have been serving and billing since July, at the same credit prices you see here.
/v1/ai/answer
A grounded answer with [n] citations back to each source.
2 with your own sources · 4 if we fetch the SERP
/v1/ai/content-brief
Angles, sections and entities to cover, from the top of the SERP.
/v1/ai/visibility
Whether and how a brand shows up in AI engine answers.
/v1/ai/serp-diff-explain
Why the rankings moved, and what to do next.
/v1/ai/competitor-brief
Who ranks, on what, and where the gaps are.
/v1/ai/review-summary
Recurring themes, sentiment and concrete actions.
/v1/ai/keyword-cluster
Groups a keyword list by topic and intent.
/v1/ai/serp-intent
Intent classification plus the SERP features that signal it.
/v1/ai/content-gap
Topics your competitors cover and you do not.
/v1/ai/paa-expand
Turns PAA questions into grounded answers.
/v1/ai/summarize
Multi-document summary with citations back to each document.
/v1/ai/rerank
Reorders documents by relevance to a query. Built for RAG pipelines.
POST /v1/ai/*Run any of them without writing code in the Playground, or read the request bodies in the full reference.
The licensed layer
Underneath every capability above sits the same thing: results obtained under licence, refracted into typed tables with a fixed color each. All six bands are live and billing today. Hover one to see the fields it carries.
POST /v1/search
Organic results as a ranked, typed table. No HTML, no XPath, no parsing.
POST /v1/news
Headlines with source and recency attached to every row.
POST /v1/maps
The local pack with ratings, review counts and stable place IDs.
POST /v1/serp/ai-overview
The answer blocks with the sources they cite — the raw material every GEO question is made of.
POST /v1/shopping
The product grid decomposed — prices, sellers, thumbnails.
POST /v1/images
Image results with originals and provenance attached.
Beside the six bands run the rest of the catalogue on the same key and the same ledger: backlinks, keyword research, on-page, business and local data, Amazon, reviews, YouTube and Trends — the parts the composed capabilities read from. See every endpoint.
Decomposition
Pick a query. The raw, unreadable SERP on the left refracts into typed, spectral JSON blocks on the right.
google_search
organic_results[0]Search{
"position": 1,
"title": "Best home EV chargers of 2026, tested",
"link": "https://www.techradar.com/best/ev-chargers",
"displayed_link": "techradar.com › best",
"snippet": "We installed 14 level-2 chargers and measured speed, app quality and reliability."
}organic_results[1]Search{
"position": 2,
"title": "EV charger buying guide: amps, connectors, smart features",
"link": "https://www.caranddriver.com/shopping-advice/ev-charger-guide",
"snippet": "A 48-amp unit adds roughly 46 miles of range per hour — check your panel first."
}news_results[0]News{
"position": 1,
"title": "New NEVI round funds 1,200 fast-charging stalls",
"source": "Reuters",
"date": "2 days ago",
"link": "https://www.reuters.com/business/energy/nevi-round"
}people_also_ask[0]AI{
"question": "Is a 48A home charger worth it over 32A?",
"snippet": "Only if your car's onboard charger accepts 11.5 kW or more."
}related_searchesSearch[
{
"query": "best ev charger 2026"
},
{
"query": "level 2 charger installation cost"
}
]Local sample data — this demo never calls the API. Run it live in the playground →
Agent-native
The interface is the envelope — the same one for raw bands and for analysis. Predictable fields, machine-checkable costs, retries that never double-spend.
Every engine returns the same JSON envelope: search_metadata plus typed result tables. No HTML, ever.
Send an Idempotency-Key header and retries are safe — replays return the original response, flagged X-Idempotent-Replay.
X-Credits-Used and X-Credits-Remaining ride on every response, so agents can budget without polling a dashboard.
npm install dataswap. It waits out the 202 handoff, retries without double-charging, and hands you typed errors — the parts you would otherwise write yourself.
Context Packs, Entity Cards and the rest return the same envelope, take the same key, and spend from the same balance as a plain search. Nothing new to authenticate, meter or reconcile.
npm install dataswap — zero dependencies, TypeScript types included. Read the docs.
HTTP/1.1 200 OK
X-Request-Id: req_01JX3F0R6GQ8
X-Credits-Used: 1
X-Credits-Remaining: 74999
{
"search_metadata": {
"status": "success",
"engine": "google_search",
"cached": false,
"latency_ms": 842,
"credits_used": 1
},
"organic_results": [
{
"position": 1,
"title": "Best home EV chargers of 2026, tested",
"link": "https://www.techradar.com/best/ev-chargers",
"snippet": "We installed and tested 14 level-2 chargers…"
}
],
"people_also_ask": [ … ]
}Catalog extraction
The same pipeline that structures a SERP, pointed at e-commerce: give it a shop and get clean product records back — titles, prices, currency, availability.
POST /v1/extract/catalogcurl -X POST \
https://api.dataswap.io/v1/extract/catalog \
-H "Authorization: Bearer sk_live_…" \
-H "Content-Type: application/json" \
-d '{
"url": "https://example-store.com/",
"max_products": 50
}'
# 202 Accepted — work continues server-side
{
"job_id": "3c9a1f28-…",
"status": "queued",
"estimated_credits": 50
}GET /v1/extract/catalog/:jobId{
"status": "completed",
"products_found": 128,
"products_extracted": 50,
"credits_charged": 50,
"result": {
"site": "example-store.com",
"products": [
{
"url": "https://example-store.com/p/trail-runner-gtx",
"title": "Trail Runner GTX",
"price": "129.90",
"currency": "EUR",
"availability": "in_stock"
}
]
}
}Point it at a store, not at a list of pages. It reads the sitemap and falls back to the links on the page you submit.
Your ceiling is reserved up front, then refunded in full — you pay for the products actually extracted. A job that finds nothing costs nothing.
You get a job id straight away and poll it. Credits stay reserved, never debited, until there is a result to hand you.
Try it without writing any code in the Playground, or read the full reference.
Provenance
A signed manifest of where the data came from, when, and whether any inference touched it, travelling with the response. Verification is public and needs no API key — because the person who has to approve you does not have one.
GET /v1/provenance/public-key{
"v": 1,
"source_class": "licensed_provider",
"bot_evasion": false,
"inference": "eu-only",
"operation": "answer_landscape",
"request_id": "req_01JX3F0R6GQ8",
"retrieved_at": "2026-08-02T09:14:27.318Z",
"payload_sha256": "9f2c1ab5e0d47c3a8b6e5f01d2c7a934be8150f2c6d3e9a17b04c8f5a2e6d310",
"signature": "kR7pQ2x9Lm4vTf8sJb1cN6yWzA0hE3dG…",
"key_id": "3a91f0c47be25d18"
}Sample receipt: the shape and the signing scheme are the ones implemented, the signature above is not real. Receipts begin riding on responses together with the capabilities at the top of this page.
That this API issued that exact payload, at that time, under the policy stated in the receipt — the hash binds the signature to the bytes you hold, so an edited response no longer verifies.
That the policy itself is true. This is a signed self-attestation, not a third-party audit or certification: it proves we said it and cannot take it back, and the licence behind it is a contract you are welcome to ask to see.
Send the receipt and, if you want the strong check, the payload it covers. The endpoint is anonymous by design: an auditor with no key can confirm a signature they already hold, and nothing about your account is exposed by doing so. The public key is published at /.well-known/dataswap-provenance.json, so the check also works offline.
# Public and keyless: whoever audits you has no API key.
curl -X POST https://api.dataswap.io/v1/provenance/verify \
-d '{ "receipt": { … }, "payload": { … } }'
{
"valid": true,
"signature_valid": true,
"payload_matches": true
}Trust
If your pipeline runs on data of uncertain provenance, that risk is in your product. Ours removes it at the source — and hands you a signed receipt that says so.
We never ship anti-bot evasion, CAPTCHA solving or scraping of sources that forbid it. That is a contractual position, not a feature flag — which is why it can be signed into a receipt.
Synthesis runs on models we host on our own hardware. Your queries and the pages we fetch to ground them are not sent to a third-party model vendor, so there is no extra sub-processor in your chain.
Data minimization by default, documented flows, deletion on request, and a signed data processing agreement with every paid plan.
No SOC 2, no third-party certification, no uptime SLA yet. We would rather you find that out here than in procurement.
Pricing
One reservoir of credits covers everything: raw bands, the twelve inference tools, and the composed capabilities. A fresh search is 1 credit; cached costs half, live costs 1.5×, maps costs 2× base.
$0/mo
No included credits
Account + API key. Credits sold separately.
No card required to sign up
Prices exclude VAT. Portuguese VAT of 23% applies to customers in Portugal; for other countries the rate of your location applies, and EU businesses with a valid VAT number are reverse-charged at 0%. The exact amount is shown at checkout.
The synthesis runs on hardware we own, so there is no token meter on top of any of them — what moves the number is the licensed data a call has to fetch. The six composed capabilities above quote a ceiling before the call and settle on what actually ran: you are never billed for parts that did not execute, and never billed more than the ceiling we reserved.
Every response says what it cost. X-Credits-Used is what you actually paid; when the ceiling we held was higher than that, it comes back as X-Credits-Reserved and the difference is already in your balance by the time the response reaches you. GET /v1/pricing marks every operation fixed or variable, and says what a variable one depends on — so you can budget from the API instead of from this page.
Dial in your query volume and freshness — we point you at the cheapest plan that covers it.
Assumes /v1/search at 1 credit per query × freshness multiplier (cached ×0.5 · fresh ×1 · live ×1.5). Checkout is open — subscribe to a plan or buy a prepaid pack. Amounts exclude VAT; the exact total is shown at checkout.
Create an account and get your API key — no card required. Start on the six live bands and the twelve inference tools; the eight new capabilities land on the same key as they ship.