Skip to main content

​
Web & Media

Beyond the full browser, BOR has focused tools for searching the web, reading pages, viewing images, and generating images. These power sourced answers, research, and visual output. They live in server/protocol/handlers/tools.js (with server/tools/web-search.js, web-fetch.js, images.js).

​
Web search & fetch

ToolPurpose
web_searchSearch the web; returns ranked results (source, title, snippet).
web_fetchFetch and read a specific URL’s content.
These are how BOR gives sourced answers and researches before building. When you ask “build me a calorie tracker app”, BOR first searches and reads what good calorie trackers do, then builds the complete workflow rather than a dumb form. In the bubble, web_search renders a web card showing the sources being read. BOR is told not to dump entire pages back — it extracts the IDs, facts, or URLs it needs and moves on.

​
Use cases

  • “What’s $AAPL doing today?” → finance sources, then a concise <say> summary (a shortcut or app only if you want it persistent).
  • “Plan a weekend hike near me.” → search trails + weather, fetch details, answer with sources.
  • “Research competitors for my SaaS.” → search + fetch, synthesize into a brief (often via the marketing-research skill).

​
Images

ToolPurpose
generate_imageGenerate an image with the configured provider; returns a /api/images/… URL.
view_imageRead/understand an external image by URL.
create_svgPlace a vector graphic (used for decorative OS items, wallpaper elements).
generate_image is provider-backed (the same provider onboarding configured, if it supports image generation — e.g. for wallpapers via set_theme from_prompt). Generated images are saved under os-data/.../images/ and shown as an image card in the bubble. The footer offers actions like “Use as wallpaper”, “Save”, “Insert”, “Regenerate”. view_image lets BOR look at an external image URL and reason about it (useful when you share a link to a picture rather than attaching a file).

​
Use cases

  • “Generate a hero image for my landing page.” → generate_image, then insert the URL into the project.
  • “Make me an aqua wallpaper.” → set_theme with <wallpaper from_prompt="…">, which generates and applies it.
  • “What’s in this image?” (with a URL) → view_image and describe it.

​
Display tools

BOR can surface media it found or made directly in the bubble:
  • display_media — show media URLs BOR chose (e.g. images/video it scraped or generated). BOR picks the URLs explicitly rather than auto-previewing every URL a tool returns.
  • display_tweet — render a tweet/embed.
These keep the bubble clean: search and command output don’t auto-embed media — BOR decides what’s worth showing.

​
How media reaches the model

When a tool produces an image the model needs to reason about (a screenshot, a captured screen, a generated image), it’s attached to the model’s next turn as an image block (llmMedia). That’s how BOR can “look at” what it just produced and decide what to do next — the same mechanism the browser and computer use tools use.