Built a MAGI-powered internal knowledge base for our team

[dannelybraum77]

Just wanted to share a quick project. I set up an internal knowledge base using MAGI for our ops team's runbooks and incident documentation.

Setup:

• All runbooks converted to .mda format
• Front matter includes tags, audience: "ops-team", and purpose: "runbook"
• ai-script blocks configured to generate incident summaries and extract action items
• Footnote relationships link related runbooks (e.g., "database failover" -> "connection pool reset")

Results:

• New team members can ask our internal chatbot questions and get contextually relevant answers because MAGI metadata helps the RAG pipeline rank and filter results
• The relationship graph between runbooks surfaces related procedures automatically
• ai-script blocks auto-generate TL;DR summaries for each runbook

The purpose and audience fields in the front matter turned out to be surprisingly useful for filtering results by team role.

[chinemehinojo36]

This is great. How did you handle the conversion of existing runbooks to .mda? Was it manual or did you use url2mda?

[dannelybraum77]

Mix of both. For web-based docs, url2mda worked well as a starting point. For internal wiki pages and Confluence docs, I wrote a quick Python script to extract content and generate the YAML front matter from existing page metadata (tags, author, last-updated, etc.).

The ai-script blocks I added manually since they needed to be tailored per runbook type. But honestly, it only took a few hours for ~50 runbooks.

[catheleenkour29]

This is a solid use case. The audience field for role-based filtering is something I hadn't considered. We have different documentation needs across engineering, sales, and support teams - MAGI could help us serve the right context to each group without maintaining separate doc sets.

Would you be open to sharing a sanitized example of your runbook template?

[dannelybraum77]

Sure! Here's the skeleton we use:

---
doc-id: "runbook-<service-name>"
title: "<Service> - <Scenario>"
tags: ["runbook", "<service>", "<severity>"]
purpose: "runbook"
audience: "ops-team"
updated-date: "2026-03-15T10:00:00Z"
entities: ["<service>", "<dependencies>"]
---

Then standard markdown for steps, with an ai-script block at the end for summary generation. Works well with any RAG setup.