Skip to content

Browser Evidence Capture (Browser + Ollama)

Goal: open a real page, extract key facts, and save a screenshot as evidence.

This is a good first “browser + LLM” workflow.

Prereqs

  • Ollama running locally (ollama serve)
  • A Chromium-based browser installed (Chrome/Chromium/Edge)

Run (GUI mode)

1
2
3
export LOOPFORGE_BROWSER_HEADLESS=0

loopforge agent run --workspace . --prompt "Use browser tools to open https://en.wikipedia.org/wiki/Large_language_model . Wait for #firstHeading (browser_wait). Run browser_run_js to extract document.title and the heading text. Scroll down by 800px (browser_scroll). Read the page and write notes/browser_evidence.md with: URL, title, heading, and a 5-bullet summary based only on the page text. Save a screenshot to .loopforge/browser/wikipedia_llm.png. Close the browser."

1
2
3
$env:LOOPFORGE_BROWSER_HEADLESS = "0"

loopforge agent run --workspace . --prompt "Use browser tools to open https://en.wikipedia.org/wiki/Large_language_model . Wait for #firstHeading (browser_wait). Run browser_run_js to extract document.title and the heading text. Scroll down by 800px (browser_scroll). Read the page and write notes/browser_evidence.md with: URL, title, heading, and a 5-bullet summary based only on the page text. Save a screenshot to .loopforge/browser/wikipedia_llm.png. Close the browser."

What to expect

  • notes/browser_evidence.md
  • .loopforge/browser/wikipedia_llm.png

If the browser doesn't appear

Browser tools are headless by default. Make sure LOOPFORGE_BROWSER_HEADLESS=0 (or pass headless=false on the first browser_navigate call).