How it works — Browser Use
The web as a tool, not a distraction
NuBench AI launches a fresh, ephemeral Chromium browser session for each conversation turn that requires web access. No browser extension to install, no manual copy-paste, no screen-sharing setup. You describe what you need, and NuBench reads, navigates, and extracts on your behalf.
What NuBench can do in a browser session
Ask NuBench to open any URL and read the page contents. NuBench renders the full page — including JavaScript-heavy sites — and extracts the information you need.
Useful for pulling public pricing pages, reading competitor documentation, checking regulatory updates, or retrieving structured data from web portals that do not expose an API.
NuBench captures a screenshot of any page or element and describes what it sees. Use this to verify page layouts, capture UI states for documentation, or get NuBench to read image-based content.
Screenshots are ephemeral — they exist for the duration of the conversation turn and are not stored. PHI-classified conversations block access to external URLs by default.
NuBench can fill form fields, click buttons, and navigate multi-step web flows — all from a single instruction in the chat interface.
Form automation is scoped to what a logged-in user session can reach. Credentials used for site authentication can be stored in the NuBench Secrets Vault and referenced by name, so raw passwords never appear in the conversation.
Ask NuBench to scrape a table, a list, or any repeated structure from a web page and return it in a format you can paste into a spreadsheet, feed into another prompt, or export as JSON.
Extraction is done in the ephemeral Chromium session — no persistent scraping infrastructure, no rate-limit exposure for your IP, and no data stored between turns.
Use cases
Competitive research
Monitor competitor pricing pages, product announcements, or job postings. Ask NuBench to read five competitor sites and return a structured comparison — without a browser tab or a copy-paste session.
Regulatory monitoring
Pull the current version of a regulatory guidance page, compare it to the version from last month, and surface the changes that affect your practice — in a single conversation.
Form and portal automation
Submit structured data to web portals that lack an API. NuBench handles the navigation, field filling, and confirmation — supervised by the user who can review before any submission is made.
Data extraction for analysis
Extract tables from web pages, government databases, or public filings and pipe the structured output directly into a NuBench data analysis conversation.
Security model
- Ephemeral sessions
- Every browser session is launched fresh for each conversation turn and torn down immediately after. There is no persistent browser state, no cookies stored between turns, and no cross-conversation browser history.
- PHI classification restriction
- Conversations classified at the PHI tier block access to external URLs by default. Browser use in PHI conversations is restricted to an admin-configured allow-list of approved internal URLs. This prevents inadvertent transmission of regulated data to external sites.
- Credential handling
- Site authentication credentials can be stored in the NuBench Secrets Vault and referenced by name in browser instructions. The raw credential is resolved at dispatch time and injected directly into the browser session — it never appears in the conversation record.
- Audit trail
- Every browser action — URL navigated, form submitted, screenshot taken — is recorded in the immutable NuBench audit trail with the user identity, conversation ID, classification tier, and timestamp.
- Network isolation
- Browser sessions run inside the NuBench tenant network boundary. Egress is subject to your tenant egress policy. The Chromium instance has no access to your local network or internal services not explicitly configured as reachable.