Web fetch

⚠️

Preview — this feature is being built; the design below is the spec.

The built-in syntic_web_fetch tool lets the model read the full content of a web page or online PDF during a request. Syntic downloads the page, turns it into clean text and hands it to the model. Pair it with web search (search finds pages, fetch reads them in full) or use it alone when the user gives an address.

When to use it

“Summarise this article” or “What does this page say about returns?” with a link.
Reading documentation pages, release notes or public PDFs.
Following up on search results that need more than a snippet.

Endpoint and auth

POST https://app.syntic.ai/v1/chat/completions (and Messages), Authorization: Bearer sk_live_....

Tool definition

Field	Type	Required	Details
`type`	string	Yes	`"syntic_web_fetch"`
`max_uses`	integer	No	Fetches allowed in this request, 1 to 20. Default 5.
`allowed_domains` / `blocked_domains`	array of strings	No	As on web search.
`max_content_tokens`	integer	No	Truncate each page to this many tokens. Default 20,000, maximum 100,000.
`citations`	object	No	`{"enabled": true}` to cite passages from fetched pages.

Which addresses can be fetched

To stop the tool being used to leak data, the model may fetch only addresses that already appear in the conversation (in a user message, a tool result or an earlier search result). It cannot invent new addresses or build them from private data.

Allowed	Not allowed
Public `http` and `https` pages and PDFs	Private, internal or loopback addresses
Redirects up to 5 hops (each hop checked again)	Non-standard ports other than 80, 443, 8080, 8443
Pages up to 10 MB	Pages that need a login, cookies or JavaScript to show content

Response

Like web search: the answer text, annotations when citations are on, and a record of the fetches:

{
  "message": {
    "role": "assistant",
    "content": "The returns window is 30 days and plants must be unopened.",
    "server_tool_calls": [
      { "id": "srv_02", "type": "syntic_web_fetch", "url": "https://shop.example.com/returns", "status": "ok", "content_tokens": 1830, "truncated": false }
    ]
  },
  "usage": { "prompt_tokens": 2410, "completion_tokens": 19, "total_tokens": 2429, "server_tool_use": { "web_fetch_requests": 1 } }
}

`status`	Meaning
`ok`	Content read.
`not_allowed`	Address not in the conversation, blocked by domain rules or not public.
`failed`	Timeout, HTTP error or unsupported content.
`too_large`	Over 10 MB.

Failed fetches are reported to the model, which explains the problem or tries another source.

Examples

curl https://app.syntic.ai/v1/chat/completions \
  -H "Authorization: Bearer $SYNTIC_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "Amara S1.2",
    "stream": false,
    "tools": [{ "type": "syntic_web_fetch", "max_uses": 2, "citations": { "enabled": true } }],
    "messages": [{ "role": "user", "content": "Summarise https://shop.example.com/returns in three bullet points." }]
  }'

const data = await (await fetch('https://app.syntic.ai/v1/chat/completions', {
  method: 'POST',
  headers: { Authorization: `Bearer ${process.env.SYNTIC_API_KEY}`, 'Content-Type': 'application/json' },
  body: JSON.stringify({
    model: 'Amara S1.2',
    stream: false,
    tools: [
      { type: 'syntic_web_search', max_uses: 2 },
      { type: 'syntic_web_fetch', max_uses: 3, max_content_tokens: 15000 },
    ],
    messages: [{ role: 'user', content: 'Find the latest release notes for our CRM vendor and list breaking changes.' }],
  }),
})).json()
console.log(data.choices[0].message.content)

import os, requests
 
data = requests.post(
    "https://app.syntic.ai/v1/chat/completions",
    headers={"Authorization": f"Bearer {os.environ['SYNTIC_API_KEY']}"},
    json={
        "model": "Amara S1.2",
        "stream": False,
        "tools": [{"type": "syntic_web_fetch", "blocked_domains": ["example-competitor.com"]}],
        "messages": [{"role": "user", "content": "What does https://shop.example.com/returns say about plants?"}],
    },
    timeout=120,
).json()
for call in data["choices"][0]["message"].get("server_tool_calls", []):
    print(call["url"], call["status"])

Errors

Unified envelope. Individual fetch failures do not fail the request.

Status	`code`	Cause
400	`server_tool_not_supported`	The model cannot use web fetch.
400	`invalid_domain_filter`	Both lists sent or malformed domains.
400	`invalid_max_content_tokens`	Above 100,000.
403	`tool_disabled`	Turned off for your organisation.

Pricing and limits


Fetch fee	None. You pay only for the fetched content as input tokens.
Fetches per request	20
Timeout	15 seconds per fetch
Cache	Pages are cached for up to 15 minutes per organisation.

Build notes for the platform team

Same server-side loop as web search; ship right after it.
Every fetch and every redirect hop goes through the shared outbound URL checks with DNS pinning.
Enforce the “address already in the conversation” rule in the server, not in the prompt.

Web search Code execution

Was this page helpful?