Skip to main content
Can my agent read the article I shared? Only if you turn link understanding on — it is off by default. When enabled, Comis fetches the URL, extracts the readable content (headings + body text, no nav/ads), and hands it to the agent alongside the user’s message. This is separate from the client-side link previews that Telegram, Discord, and Slack render automatically (those are generated by the chat platform itself and your agent cannot read them without link understanding). WhatsApp, Signal, LINE, iMessage, IRC, and Email do not show client-side previews at all — link understanding is the only way the agent learns what the URL points to on those channels. When someone shares a URL in a message, Comis can automatically fetch the web page and extract its readable content. Your agent then sees both the original message and a summary of what the linked page says. This gives your agent context about shared links without the user having to copy-paste the page content.
You don’t need to understand the technical details to use this feature. The configuration examples below are copy-paste ready.
1

URLs are detected

Comis scans incoming messages for URLs. Standard web links starting with http:// or https:// are recognized. Up to 3 links per message are processed by default (configurable).
2

Pages are fetched safely

Each URL is fetched with SSRF protection — Comis will not fetch internal or private network addresses. A readable extraction pulls the main content from the page, stripping navigation menus, advertisements, sidebars, and scripts. The result is clean, readable text.
3

Content is injected

The extracted text (up to 5,000 characters per link) is added to the message context so your agent can reference it. Your agent sees the user’s original message plus the content of each linked page.
Link understanding is disabled by default. You need to explicitly enable it in your configuration before it will process any URLs.
Link understanding is most useful when your users regularly share web content and expect the agent to discuss it. Common scenarios include:
  • News and research groups — Users share articles and want the agent to summarize or comment on the content.
  • Support channels — Users share documentation links and the agent needs to reference the linked page to answer questions.
  • Content review — Users share blog posts, product pages, or forum threads for the agent to analyze.
If your agent rarely needs to understand web pages, you can leave link understanding disabled to avoid unnecessary network requests.

​
Security

Link fetching includes built-in SSRF (Server-Side Request Forgery) protection. SSRF is an attack where a malicious user tricks a server into fetching internal resources that should not be accessible from outside. Comis prevents this by refusing to fetch URLs that point to:
  • Private IP addresses (10.x.x.x, 192.168.x.x, 127.x.x.x)
  • Internal network hostnames
  • Non-HTTP/HTTPS protocols
This prevents malicious users from tricking your agent into accessing internal services on your network. For more information on Comis security layers, see Defense in Depth.
The SSRF protection runs automatically on every URL before any content is fetched. You do not need to configure or maintain a blocklist — Comis handles it for you.

​
What the Agent Sees

When link understanding is enabled and a user sends a message containing a URL, your agent receives the original message text plus the extracted page content. The content is formatted as readable text with the page title, so the agent knows which link each block of content came from. If a page cannot be fetched (timeout, 404 error, blocked by the site), Comis silently skips it. The agent still sees the original message with the URL — it just does not get the extracted content for that link.

​
Limits

These limits control how much content is fetched and how long Comis waits for each page to respond.
SettingDefaultDescription
maxLinks3Maximum URLs processed per message
fetchTimeoutMs10,000 msTimeout per URL fetch
maxContentChars5,000Maximum characters extracted per link
If a message contains more URLs than maxLinks, only the first URLs (in order of appearance) are processed. The rest are ignored. If a page’s readable content exceeds maxContentChars, the text is truncated. This keeps the agent’s context window from being overwhelmed by very long pages.

​
Configuration

integrations:
  media:
    linkUnderstanding:
      enabled: false                    # Must be explicitly enabled
      maxLinks: 3                       # Max URLs per message
      fetchTimeoutMs: 10000             # Timeout per fetch (10 sec)
      maxContentChars: 5000             # Max chars per link
      userAgentString: "Comis/1.0 (Link Understanding)"  # Outbound User-Agent
To enable link understanding, set enabled: true: 
integrations:
  media:
    linkUnderstanding:
      enabled: true
All other settings are optional and have sensible defaults.
Start with the defaults and only adjust limits if you find that extracted content is being cut off or fetches are timing out on slower sites. Increasing maxContentChars gives the agent more page content to work with but uses more of the agent’s context window.

​
What Happens When Disabled

When link understanding is disabled (the default), URLs in messages are passed through to the agent as plain text. The agent sees the URL but does not know what the page contains. Users would need to copy-paste the relevant content from the page if they want the agent to discuss it.

Media Overview

Back to media overview

Security

SSRF protection and other security layers

Config Reference

Full configuration reference

Vision

Image and video analysis