Which API should an AI agent use when it already knows the source URL?

Use URL to Markdown when compact readable text is enough, or Web Extract when structured extraction is the real job. Use Web Research only when the source still needs to be found.

When should an agent use a stock quote versus market-data verification?

Use Quick Quote when one latest provider-backed reading is sufficient. Use Truth Verify when the downstream decision needs multiple provider readings and explicit disagreement evidence.

What should an agent check before paying a provider?

Inspect provider identity, runtime evidence, payment terms, limitations and the fresh target challenge. A payment challenge is not proof of settlement or accepted delivery.

AI AGENT API BUYER GUIDE · JOB FIRST · PROOF BEFORE PAYMENT

Choose the smallest correct API for the agent’s actual job.

The fastest way to waste agent budget is to buy a broad platform for a narrow task—or to treat a cheap single reading as verified evidence. This guide routes seven common jobs by what the caller actually needs, what can be checked before spend, and what each product deliberately does not claim.

Decision rule: known source → smallest extraction job; unknown source → research; one market reading → quote; corroboration → verification; provider uncertainty → trust check; currency normalization → reference-rate conversion. Always inspect current runtime terms before payment.

1. I already know the URL. I need readable model context.

URL to Markdown API is the narrow choice: one public HTTPS URL in, compact text/markdown out after accepted delivery.

Best when: known source, RAG/context ingestion, no separate scraper account, pay-per-read economics.

Do not choose it when: JavaScript execution, authenticated pages, clicks/forms, whole-site crawling, or advanced browser anti-bot handling is required. Those belong on a browser/rendering platform.

2. I know the URL, but I need structured extraction.

Web Scraping API for AI Agents is the structured one-page extraction job. It is distinct from the simpler Markdown reader.

Best when: the agent needs clean structured public-web extraction rather than only Markdown.

Do not choose it as a proxy fleet, private-network fetcher, authenticated browser session or whole-site crawler.

3. I do not know the source yet.

Web Research API combines live search with safe source extraction so the result is grounded in discoverable sources instead of synthetic URLs.

Best when: the question is “find evidence,” not merely “read this URL.”

If the source is already known, route down to URL-to-Markdown or Web Extract instead of paying for search you do not need.

4. I need one stock reading.

Stock Quote API / Quick Quote asks for one US equity ticker and returns the latest available real provider-backed price and volume reading after accepted delivery.

Choose it for: lightweight checks, alerts and agent routing.

It is not a brokerage feed, execution API, or guarantee of exchange-native realtime data.

5. I need stronger market evidence.

Truth Verify / Fact Verification API is for corroboration: available independent provider readings, disagreement metrics and proof metadata.

Choose it when: the downstream action deserves multiple readings rather than one quote.

It verifies ticker market-data evidence. It is not a universal internet fact-checking oracle.

6. I am unsure whether to trust the provider itself.

Provider Trust moves the question up one level: inspect identity, runtime and payment-surface evidence before an unfamiliar provider is allowed into a tool/spend policy.

Choose it before connection or spend when the source/provider is the uncertainty.

Evidence reduces uncertainty; it is not a universal safety certification. Missing evidence stays unknown.

7. I need to normalize money values across currencies.

Currency Conversion API / FX Convert uses European Central Bank reference-rate data for deterministic pricing/reporting normalization.

Choose it for: budgets, vendor-price comparison, reporting and machine-commerce normalization.

Reference-rate calculation is not executable bank/broker FX and does not move assets.

Where SML deliberately competes differently

Many conventional APIs optimize around accounts, API keys, monthly plans or broad platform bundles. SML’s narrow x402 jobs optimize around a different machine-commerce workflow: inspect first → receive fresh terms → caller applies spend policy → authorize → verify settlement → accept delivery.

That difference is only useful when the narrow SML job actually fits. A free API, browser-rendering platform, full market-data subscription or general-purpose fact checker can be the better tool when its capabilities match the task better. The routing rule is correctness before capture.

Payment and failure truth

Canonical SqueezeOS paid surfaces currently use x402 v2, Base mainnet eip155:8453, Base USDC and canonical payTo 0xc29185fa176357612f3194735753e520e91adc46. Exact terms must come from the fresh target PAYMENT-REQUIRED challenge.

A 402 challenge is not a purchase. A signature is authorization, not settlement. Where a route declares a no-charge failure boundary, missing real evidence should not be converted into a fabricated paid response.

Machine-readable decision matrix

Agents can retrieve the same routing logic directly from /.well-known/agent-api-buyers-guide.json, then resolve exact endpoint truth from the linked live contract/OpenAPI.