Skip to main content
The Web Fetch tool lets Neo Agent retrieve the content of a specific public web page — Neo’s own docs, Microsoft Learn, or vendor knowledge bases — and read it inline as part of resolving a ticket or answering a question.

​
What It Does

When enabled, the agent can:
  • Pull a specific page from an allowlisted host by URL
  • Read product documentation, vendor KBs, and configuration guides
  • Pair with Web Search — search to find the URL, then fetch to read it
HTML pages are automatically converted to clean Markdown so the agent gets only the content (no page chrome, scripts, or styles). Plain text and Markdown pages pass through unchanged.

​
Allowlist

For safety, only a curated set of public documentation hosts is reachable:
  • Neo: docs.neoagent.io
  • Microsoft: learn.microsoft.com, docs.microsoft.com, developer.microsoft.com, support.microsoft.com
  • PSA vendors: developer.connectwise.com, halopsa.com, webservices.autotask.net, developer.servicenow.com
  • RMM and doc vendors: docs.ninjaone.com, rmm.datto.com, support.huduapi.com, kb.itglue.com
You can add extra hosts per agent under the tool’s settings (e.g. your own runbook site). Anything not on the allowlist — including internal IPs, private networks, and unrecognized hosts — is rejected. The allowlist also governs the agent’s sandbox: scripts it runs there can reach exactly these hosts and nothing else. That lets an agent pull a large or paginated data source itself — an open data feed, a status API — and process it in the sandbox in one step. Agents without the Web Fetch tool have no outbound web access from the sandbox at all.

​
How to Use

  1. Enable the Web Fetch tool on the agent.
  2. (Optional) Add additional hosts under the tool settings for this agent.
  3. The agent uses Web Fetch when it needs the full content of a specific page — typically after a Web Search returns a relevant URL.

​
Large Pages

A fetch returns the whole page in a single call. When a page is too large to fit in the agent’s working context, Neo saves it to a file in the agent’s sandbox and gives the agent a preview plus the file path; the agent then processes it there — parsing the JSON, filtering rows, extracting the part it needs. Nothing is silently dropped, and a large page costs one fetch rather than one per chunk.

​
Limits

  • HTTPS only.
  • Pages up to 5 MB. Beyond that the response is truncated, and the agent is told so.
  • 10-second total timeout.