Use built-in browser and web research tools

Agents can research public information and inspect interactive pages with Helpin's built-in tools. Choose the smallest tool set needed for the question. These are agent capabilities, not a general-purpose browsing session shared with every agent. The tool catalog is the current reference for available names and inputs.

Choose a research tool

Need	Tool	Use it for
Find public sources	web_search	Discover relevant pages and recent information. Open the source before treating a snippet as definitive.
Read a known public URL	fetch_url	Extract text from a credential-free page without browser interaction.
Discover related public pages	crawl_url	Follow a bounded set of same-host links, such as changelogs or release notes.
Inspect an interactive page	browser_open, browser_snapshot, browser_act	Open an allowed page, locate controls and perform bounded clicks, form actions or scrolling.
Capture UI evidence	browser_screenshot, browser_record	Save a screenshot or bounded screencast of the current browser session.

Use web tools for public pages that do not require interaction. Use the browser when client rendering or clicks matter. A browser session is temporary and closes at the end of the run, so do not assume a sign-in persists to another run. Browser access depends on the run's permitted host scope and availability in your edition.

Capture your app with the browser

To ask an agent for accurate product screenshots or a screencast, provide the full URL of the dev or staging environment, the route to capture, the desired UI state, and the version or commit deployed there. If the app requires sign-in, provide a dedicated test account's credentials through an approved private channel or a secret input when the run asks. Do not put passwords in the article, repository, public URL, screenshot, or recorded video. If SSO, MFA, a network allowlist or a CAPTCHA blocks automated login, complete that step with a person or use a safe preauthenticated environment permitted by your organization. The URL must be reachable under the run's browser host policy.

For example, ask: Open our staging app at the URL I provide, sign in with the test account I supply privately, navigate to the Team settings page, capture the members table at desktop width and record the invite flow. Use sample data only. Return both captures in this chat, then embed the approved ones in the Help Center document I name. Do not include actual credentials in that prompt if the channel is not approved for secrets.

1. The agent opens the allowed URL with browser_open. It reads the interactive snapshot and uses browser_act to fill and submit the login form, then navigates to the requested screen. After navigation, it refreshes the snapshot before reusing element references.

2. The agent checks the environment, deployed version, viewport, sample data and UI state. If it cannot sign in or the target UI is absent, it reports the blocker rather than inventing a capture.

3. For a still image, it calls browser_screenshot on the page or relevant element. For a demo, it calls browser_record to start, performs the steps with browser_act, then calls browser_record to stop. Recording produces a private MP4 artifact.

4. The agent can show or link the resulting private artifacts in the run chat for review. Only share the approved result in public Docs. Inspect each image or recording for secrets, notifications, real customer data and stale UI before using it.

The browser session is temporary. Sign-in normally needs to be repeated in another run. For public research that needs no clicks or login, use web_search, fetch_url or crawl_url instead.

Put captures in Docs

A browser capture is a private run artifact. To embed one in Docs, first create or edit the document text, then have the agent call insert_document_artifact with the document ID and the exact artifact_id returned by browser_screenshot or the stopped browser_record. Add a description and, where useful, a caption and placement anchor. Confirm that insertion succeeded and review the rendered block. A filename, artifact_ref, URL or Markdown image link pasted into document text does not embed the capture. A chat artifact is available for review, but it is not automatically in a document. For a manual image, use the Docs image picker or paste; for a hosted screencast, use a supported video embed such as Loom, Vimeo, YouTube or Wistia. Check public playback and article rendering before publishing.

Grant access deliberately

In Automation → Agents, edit a custom agent's Capabilities and select only the tools it needs. A tool in the catalog does not grant access by itself. Write actions may require approval under the agent's run policy. Browser and coding tools are not available in the Community edition. See Built-in tools overview, Give agents tools and skills and Images and file attachments.