Skip to content

Web Tools

janito provides tools for accessing the web: fetching content from URLs, searching the web, and rendering pages with a headless browser.

All tools live in the dedicated net toolset (janito/tools/net/), which is auto-loaded, so they are available in every session without extra flags. WebSearch additionally gates itself on the brave_api_key secret (see Setup) and is only advertised to the model once that secret is configured. HeadlessBrowse gates itself on the presence of a Google Chrome (or Chromium-based) browser binary.

Setup

The WebSearch tool searches the web via the Brave Search API (GET /res/v1/web/search), querying Brave's independent search index. It is only loaded when the brave_api_key secret is set.

  1. Create an account and get a subscription token from the Brave Search API
  2. Store the token as a janito secret:
# Set your Brave Search subscription token
janito --set-secret brave_api_key=your-brave-subscription-token

That's it — WebSearch is now available. You can verify the secret is stored with:

janito --list-secrets        # shows secret names (never values)
janito --get-secret brave_api_key

See Secrets for more on the secrets store.

GetUrl

GetUrl fetches content from any http:// or https:// URL and requires no setup.

llms.txt site maps

Before fetching a site URL (a hostname or hostname/path), GetUrl looks for an llms.txt site map. It probes the two standard locations in priority order with lightweight HEAD requests (to minimize bandwidth):

  1. <origin>/llms.txt — root level
  2. <origin>/.well-known/llms.txt — well-known path

If one of them answers 200 OK, the tool fetches it with a GET request and returns its content as-is (no Markdown parsing) as a map for further exploration. The discovery probes are silent — only a successful retrieval is reported. When no llms.txt exists, the tool falls back to fetching the requested URL normally. Fetching an llms.txt URL directly never triggers a discovery loop. The probe can be disabled by passing skip_llms_txt=True, which fetches the URL as-is without checking for an llms.txt site map.

HeadlessBrowse

HeadlessBrowse renders a URL with headless Google Chrome and returns the page's DOM. Unlike GetUrl (a plain HTTP fetch), it executes JavaScript, so it sees content that only appears after the page's scripts run (SPAs, client-side rendering, etc.). It requires no setup — but it is only loaded when a Google Chrome (or Chromium-based) binary is found on the system. If Chrome is missing, the tool is simply not advertised to the model (check /tools for the skip reason).

Available Tools

Tool Description Permissions
GetUrl Fetch content from a URL r
WebSearch Search the web via the Brave Search API r
HeadlessBrowse Render a URL with headless Chrome (runs JavaScript) r

Usage

Example Prompts

# Search the web (requires the brave_api_key secret)
janito "Search the web for the latest Python release notes"

# Search and then read a result
janito "Find recent news about AI agents and summarize the top result"

# Fetch a URL directly
janito "Fetch https://example.com and summarize it"

# Render a JavaScript-heavy page with headless Chrome
janito "Browse https://example.com with headless Chrome and summarize what it shows"

GetUrl Parameters

Parameter Type Default Description
url str — The URL to fetch (must be http:// or https://). Required.
max_length int 5000 Maximum number of characters to return
max_lines int 200 Maximum number of lines to return
timeout int 10 Request timeout in seconds
follow_redirects bool True Whether to follow HTTP redirects
skip_llms_txt bool False When True, fetch the URL as-is without probing for an llms.txt site map
threshold int 10000 Content size (chars) above which the full content is written to a temporary file instead of being returned inline (never applies to llms.txt)

When fetched content exceeds threshold, it is stored in a temporary file (removed on exit) and the tool returns the file path plus a message to explore it with search tools, instead of blowing up the model context. This never applies to llms.txt site maps: they are always returned inline in full, regardless of size.

HeadlessBrowse Parameters

Parameter Type Default Description
url str — The URL to browse (must be http:// or https://). Required.
max_length int 10000 Maximum number of characters to return
max_lines int 500 Maximum number of lines to return
timeout int 30 Chrome process timeout in seconds
wait_ms int 1000 Virtual time budget (ms) to let JavaScript run before dumping the DOM
threshold int 10000 DOM size (chars) above which the full content is written to a temporary file instead of being returned inline

The wait_ms parameter controls how long the page's JavaScript is allowed to run — increase it for heavily scripted pages. As with GetUrl, oversized DOMs are stored in a temporary file (removed on exit) rather than returned inline.

WebSearch Parameters

Parameter Type Default Description
query str — The search query (max 400 characters / 50 words). Required.
count int 10 Number of web results to return, 1–20
country str — 2-letter country code for result origin (e.g. "US")
search_lang str — Language code for results (e.g. "en")
safesearch str — Adult-content filter: "off", "moderate", or "strict"
freshness str — Page-age filter: "pd" (24h), "pw" (7d), "pm" (31d), "py" (365d), or a custom "YYYY-MM-DDtoYYYY-MM-DD" range

The response is a clean, model-friendly list of web results (title, url, truncated description, age, language) plus any news hits, along with the result count and the request duration.

Troubleshooting

"Secret 'brave_api_key' is not set"

WebSearch is not loaded because the secret is missing. Run:

janito --set-secret brave_api_key=your-brave-subscription-token

"HTTP Error 401" / "HTTP Error 403"

The subscription token is invalid or rejected by Brave. Re-check the token at brave.com/search/api and set it again with janito --set-secret brave_api_key=....

"URL must start with http:// or https://"

GetUrl only supports HTTP and HTTPS URLs.

"Google Chrome (or another Chromium-based browser) was not found"

HeadlessBrowse is not loaded because no Chrome/Chromium binary was found on the system. Install Google Chrome (or Chromium, Brave, or Microsoft Edge) and restart janito. The tool finds the browser via PATH plus the standard macOS and Windows install locations.

More Info