Retrieve public web and platform evidence

Web Access uses Treg to read public pages, search the web, and retrieve structured platform data. It is a fallback workflow for tasks that existing search, browser, fetch, or installed connections cannot cover, or when the user specifically requests Treg.

The workflow chooses a job before a provider: read a known URL, search for relevant pages, retrieve a platform object through structured fields, map a site before selecting pages, extract typed fields, or use dated news search. This keeps retrieval focused on the information needed rather than collecting an entire site by default.

Use it for authorized public research when ordinary tools cannot read a page, JavaScript or rate limits obstruct a simple fetch, or a platform lacks an existing connector. Existing tools come first whenever they can complete the task.

Do not send localhost, staging, intranet pages, or content behind the user's login through the service. The workflow does not ask for browser cookies, log in as the user, evade paywalls, or bulk-collect private personal data. A provider refusal is handled through an authorized alternative or reported limitation rather than repeated attempts.

An existing Treg MCP connection or CLI can support the workflow. Missing setup is disclosed rather than installed automatically. Calls, uploads, remote changes, scheduling, and external feedback stay within their separately authorized scopes and approved budgets.

Inputs can be a public URL, a research question, a platform object, a site topic, or a requested extraction schema. Catalog discovery searches the job in plain language and inspects current endpoint details before use. Routed options are preferred where suitable, with provider hit rate, speed, pricing, and available inputs considered.

Start with one small call and inspect the serving provider, attempted routes, and reported charge before scaling. If no capability fits, report the gap locally; submitting an external catalog request needs explicit authorization.

A zero-cost cap supports search and extraction through a preferred TinyFish route. The cap is essential: it refuses paid fallback rather than silently allowing a supposedly free call to spend. The documented setup check is:

treg call treg.web.search --data '{"q":"treg web access"}' --header 'X-Treg-Route-Prefer: tinyfish' --header 'X-Treg-Route-Max-Cost: 0'

Free availability remains a catalog claim to verify, not a guarantee. A direct TinyFish multi-URL fetch is another documented option without fallback. Platform data can carry charges, which must be disclosed before a batch.

Routing headers can prefer or exclude a provider, cap one call, or stop at the first miss. Set a cap for batches because a failed or empty route can otherwise continue through providers, including ones that bill misses. Per-call caps supplement an approved total budget.

The job catalog covers markdown page reads, typed extraction, site mapping and crawling, research publications, recent news, social profiles and posts, video comments, store products, reviews, places, and Google result fields. Platform-specific jobs are preferred over scraping their HTML because they return structured data.

Research habits reduce unnecessary retrieval: keep pages as markdown, remove navigation and footers, search before reading, and keep each source URL beside the fact it supports. Example compositions include company research, topic-focused documentation reading, social listening, cross-store product comparison, review summaries, and video notes.

Page monitoring compares retained text hashes and reports differences on a later run. Recurring scheduling requires explicit approval rather than following automatically from one research request. Final results retain source links, reported costs, and limitations; feedback remains local unless submission is requested.

Treg MCP catalog_search, catalog_get, call, and balance map to CLI operations. Routed web search and extraction provide retrieval, while discovered platform providers supply structured records. Current schemas and prices take precedence over pinned examples. The imported documentation carries upstream license terms with additional restrictions and does not establish live provider availability or tested account access.