Choose the smallest web extraction job that solves the task.
SML exposes narrow web primitives rather than forcing every agent through a heavyweight browser workflow. If the source URL is already known, use Markdown reading. If you need cleaner structured text and extraction metadata, use Web Extract. If you do not know the source yet, use Web Research first.
Build a request
Ready.
Agent Web Access — $0.001
One public HTTPS URL becomes compact LLM-ready Markdown. This is the efficient choice for RAG ingestion, citation capture, summarization, and agent context when the target page is already known.
/api/web/read-markdown?url=…
Web Extract — $0.005
Use structured extraction when downstream code benefits from clean text plus extraction metadata rather than a Markdown document. It is useful for pipelines that need deterministic fields, content inspection, or normalization before another tool runs.
/api/web/extract-once?url=…
When not to use this
Do not use these endpoints for private-network targets, pages requiring caller credentials, browser automation, authenticated sessions, or JavaScript-heavy interaction. SML does not forward caller credentials, execute page JavaScript, or silently bypass explicit robots restrictions.
Failure policy is part of the product
Invalid, private, blocked, unavailable, unsupported, oversized, or empty targets should return a no-charge alternative rather than being counted as accepted delivery. Agents can inspect the live HTTP response first; a 402 means the requested deliverable passed preflight and the fresh x402 terms should be evaluated before any authorization.