Changed your mind?
Request a return within thirty days of delivery. Include your order number.
Web Foundation
Search discovers relevant pages. Fetch and OCR bring the selected page, document or scan into your workflow as readable text.
Where should I look?
A focused query. Relevant sources.
What does it say?
The selected source. Its useful text.
Discovery with control
Explore prepared search briefs. Filters narrow where you look; they do not certify that a result is correct.
Start broad enough to discover relevant sources. Separate unrelated questions rather than packing them into one query.
Country and language filters narrow discovery. They are not a substitute for checking which rules apply to the reader.
This separate US-focused brief uses the Federal Trade Commission domain, not the Brazilian context of the previous example. Check the page date, jurisdiction and exact passage before relying on it.
Prepared query and filter examples, not live searches or search results. The checklist is an application workflow, not a guaranteed response schema. Web Search supports up to 25 requested results.
Fetch / OCR
Read an accessible URL or a supported file. OCR extracts writing inside scans and images; the source still deserves review.
Request a return within thirty days of delivery. Include your order number.
Request a return within thirty days of delivery. Include your order number.
Inspect the filter every month. Replace it when damaged.
Inspect the filter every month. Replace it when damaged.
One carton requires inspection.
RECEIVEDOne carton requires inspection.
Verify names, dates and numbers against the scan. Layout and image quality affect extraction.
Fictional source previews and prepared excerpts. No live upload, fetch or OCR. Extraction provides text, not a generated answer or an indexed knowledge base.
Choose your entry point
The capabilities connect, but their input contracts are not interchangeable.
Search from your backend. Fetch accepts URLs or inline base64 data URIs in contents, up to 10 MB per item, with per-item results and errors.
Give an MCP client web_search and fetch_url. Fetch accepts one to five public URLs; use the API for inline files.
Enable WebSearch and OpenUrl during inference. Multi-step research with AdvancedWebUsage is a separate capability with different billing.
Search when you need to find it. Fetch directly when you already have it. Treat retrieved content as untrusted material, not instructions for your agent.
Search is billed per search. Extraction uses processingUnits with plan-based daily allowances. Optional schema-guided JSON uses separately billed jsonProcessingUnits, outside that allowance. Model analysis costs extra. Fetch API requires a positive balance.
Fetch reads text; Media Descriptions interprets media. Neither bypasses restricted access or guarantees perfect output.
Current billing and limits