Page access: capabilities and boundaries
Describes the capabilities and boundaries of page access for reading public web pages in a Web research Executor.
This page covers page access capabilities, session state reuse, and network and site boundaries for Web research.
- Capability boundary for public URL access and extraction.
- Session state reuse across related Web research reads.
- Network and site boundaries for public web content.
Summaries were generated by AI.
Within a Web research Executor, page access lets the Agent Runtime read one public HTTP or HTTPS address. It renders page JavaScript and returns extracted Markdown, the page title, and the final URL. Your application does not receive a browser tab, page structure, or browser object.
Page JavaScript can make normal site requests and change site-side state. A repeated page read repeats the external request. Page access does not provide a supported login workflow.
Capability boundary
- Page access does
- Open a public HTTP or HTTPS URL
- Page access does not
- Provide supported access to arbitrary protocols, local files, or private networks
- Page access does
- Render JavaScript before extraction
- Page access does not
- Give your application a live tab or page structure
- Page access does
- Report the title and final URL after navigation
- Page access does not
- Prove that the page is safe, correct, or authoritative
- Page access does
- Extract supported page or online-document content as Markdown
- Page access does not
- Preserve the source layout or return a screenshot or PDF
- Page access does
- Reuse limited site state for related reads
- Page access does not
- Click controls, complete forms, enter credentials, or run a login workflow
A URL can open but return little or no usable Markdown. Access controls, a visual-only layout, an unsupported format, a large document, or an extraction failure can cause omissions. Use another public representation if you need the exact layout or complete source file. Troubleshoot Web research gives the safe recovery paths.
Session state
Related reads within the same Web research scope can reuse cookies and origin storage set by visited pages. This supports public sites that need continuity across more than one read. It does not restore a live tab, provide durable application storage, or create a supported sign-in flow.
The saved state is scoped by Tenant, root Task, Agent Identity, and session_id, not by child Executor. Two Web
research branches inside the same root Task can therefore share state if they choose the same session_id. Give
concurrent branches distinct session identifiers when their cookies or origin storage must not mix.
See Data protection for the page-cache rule and Threat model for how external page text can affect later work.
Network and site boundaries
Page access supports public web content only. The submitted root URL is screened before navigation, but that check does not pin every later connection. Redirects, page resources, and JavaScript can open additional addresses that are not covered by the root-URL decision. Do not use Page access as a private-network isolation control. It does not provide supported access to local files, private-network destinations, or other URL schemes. See Security for the controls and residual risks of web access.
Public availability does not grant a right to collect or reuse content. Follow the publisher’s terms and the Duale AI Terms of Use.