Web Fetch Integration
The Web Fetch integration lets agents retrieve the content of any public URL on demand during a conversation. This is different from the Knowledge Base — instead of searching pre-indexed content, the agent fetches a live page at the moment it's needed.
When to use Web Fetch vs Knowledge Base
| Knowledge Base | Web Fetch | |
|---|---|---|
| How it works | Pre-indexed semantic search | Live HTTP fetch at query time |
| Best for | Your own docs, FAQs, policies | External pages, real-time lookups |
| Freshness | Updated when you re-index | Always live |
| Speed | Fast (vector search) | Slower (network call per use) |
| Requires install | No | Yes — install from Integrations |
Use Knowledge Base for your own documentation that you want the agent to search across. Use Web Fetch when the agent needs to read a specific page it doesn't know in advance — for example, a user-provided link, a live status page, or external documentation.
Installing
- Go to your project → Integrations → Browse
- Find Web Fetch and click Install
- No credentials required — click Install to confirm
- Attach the tool to your agent in the agent editor → Tools tab
Actions
fetch_page — Fetch Page
Fetches a public URL and returns the page's title and readable text content.
Tool reference: {{tool:web_fetch_fetch_page}}
Parameters:
| Parameter | Required | Description |
|---|---|---|
url | Yes | Full URL to fetch (must start with http:// or https://) |
Returns:
| Field | Description |
|---|---|
url | The URL that was fetched |
title | Page title extracted from the HTML |
content | Clean readable text (navigation, scripts, and footers removed) |
word_count | Number of words in the extracted content |
Content is capped at 12,000 characters to fit within the agent's context window.
Using Web Fetch in an agent
Attach the Web Fetch tool and instruct the agent when to use it:
If the user shares a URL and asks you to read it or answer questions about it,
use {{tool:web_fetch_fetch_page}} to fetch the page content, then answer based on what you find.
Do not make up content — only answer from what the fetch returns.
Docs-first support pattern
Combine Web Fetch with the Knowledge Base and Freshdesk for a full support flow:
When a user asks a question:
1. Search your knowledge base for relevant documentation.
2. If you find a relevant answer, share it and ask: "Does this resolve your issue?"
If not resolved and the user provides a specific docs link:
3. Use {{tool:web_fetch_fetch_page}} to read that page and try to answer from it.
If still unresolved:
4. Ask for the user's email address.
5. Use {{tool:freshdesk_support_create_ticket}} with a summary of the conversation,
what was searched, and why it wasn't resolved.
Limitations
- Only works with publicly accessible URLs — pages behind login or auth will return an error
- Returns text only — images, tables, and complex layouts may lose formatting
- Maximum 12,000 characters per fetch — very long pages are truncated
- Not suitable for APIs that return JSON — use the API Call integration for those
Troubleshooting
Page content is empty or garbled
Some pages use JavaScript rendering — the server returns an empty shell and the content loads client-side. Web Fetch reads the raw HTML, so JS-rendered pages won't work. Use the Knowledge Base URL source instead, which can handle these sites better.
Connection error
The URL may be behind a firewall, require auth, or be unreachable from the Vois AI backend. Verify the URL is publicly accessible from outside your network.