Web Fetch

Use gonkagate:web_fetch when the model should read a specific public URL. GonkaGate fetches the URL, extracts useful text from HTML, plain text, or PDF content, and passes the result back to the model.

Use Web Search when the model needs to find sources. Use Web Fetch when you already have the URL.

Quick start

export GONKAGATE_API_KEY="gp-your-api-key"

curl https://api.gonkagate.com/v1/chat/completions \
  -H "Authorization: Bearer $GONKAGATE_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "qwen/qwen3-235b-a22b-instruct-2507-fp8",
    "messages": [
      {
        "role": "user",
        "content": "Summarize the page at https://example.com/article."
      }
    ],
    "tools": [
      { "type": "gonkagate:web_fetch" }
    ]
  }'

How it works

1. Add { "type": "gonkagate:web_fetch" } to tools.
2. The model decides which URL from the conversation it needs to fetch.
3. GonkaGate fetches the public URL and extracts readable content.
4. The extracted result is returned to the model as tool output.
5. The model uses that content to answer normally in choices[0].message.content.

Configuration

{
  "tools": [
    {
      "type": "gonkagate:web_fetch",
      "parameters": {
        "max_uses": 5,
        "max_content_tokens": 20000,
        "allowed_domains": ["docs.example.com"],
        "blocked_domains": ["private.example.com"]
      }
    }
  ]
}

Domain filtering

Use allowed_domains when the request should stay inside a known docs site or product domain.

{
  "tools": [
    {
      "type": "gonkagate:web_fetch",
      "parameters": {
        "allowed_domains": ["docs.gonkagate.com", "github.com"]
      }
    }
  ]
}

Domains are matched against the exact domain and subdomains. For example, example.com also allows docs.example.com.

Supported URLs and content

• URLs must be absolute http:// or https:// URLs.
• URLs with embedded credentials are rejected.
• Private, local, or non-public network targets are rejected.
• HTML, plain text, and PDFs are supported.
• Very large or unsupported responses can fail before the model receives content.
• stream: true is not supported with gonkagate:web_fetch.
• If you also provide user-defined function tools, the model must not call Web Fetch and your function tool in the same assistant turn.

Tool result

The model receives a tool result like this. Your application normally sees only the final model answer, not this raw object, unless the model includes details from it.

{
  "url": "https://example.com/article",
  "title": "Article title",
  "content": "Extracted page text...",
  "content_type": "text/html; charset=utf-8",
  "http_status": 200,
  "status": "completed",
  "retrieved_at": "2026-06-29T12:00:00.000Z"
}

If a fetch fails, the model receives a failed tool result:

{
  "url": "https://example.com/missing",
  "status": "failed",
  "error": "HTTP 404: Not Found",
  "retrieved_at": "2026-06-29T12:00:00.000Z"
}

Usage

When Web Fetch runs, the response can report fetch count in usage.server_tool_use.web_fetch_requests. Fetched content is also part of the model context, so it can affect normal token usage.

Common errors

See also

• Server Tools Overview
• Web Search when the model should find sources first.
• PDF Inputs Plugin when your request includes PDF files directly.
• Chat Completions API reference for the full request schema.