Skip to content

How This Manual Is Organized

Diátaxis: Guide · Audience: all readers ← Back to contents

This manual is divided into four parts, matching the three ways OPENMARU APM is used -- learning it (tutorials), looking something up when you get stuck (how-to guides), and checking an exact value (reference). It is not a book to be read from front to back but a structure for going straight to the part you need now.

What the Four Parts Are For​

PartChaptersCharacterWhen to read it
Part 1. Understanding the ConceptsE1 to E3Why and what (Explanation)When you want the big picture before opening a screen
Part 2. TutorialsT1 to T3Practice from start to finish (Tutorial)On your first day -- follow it in order and you get a result
Part 3. How-to GuidesH1 to H20"How do I do this" (How-to)When a particular task or problem comes up in production; open only that chapter
Part 4. ReferenceR1 to R10A dictionary to look things up in (Reference)When checking the exact meaning and thresholds of a chart, event, or term

Suggested Paths​

Notation Conventions​

  • Menu paths -- steps are joined with ▸, as in left menu ▸ WAS ▸ Dashboard.
  • Bold -- the names of buttons, menus, and items as they actually appear on the screen.
  • Callouts -- the boxed Note (worth knowing), Key point (remember this), and Caution (easy to get wrong) in the text are the main points of that section.
  • The head of each chapter shows its character (concept / tutorial / how-to / reference) and its audience.

Note All screen captures are in the dark theme. The positions and behaviour are the same in the light theme.

About the CogentAI Feature​

CogentAI, which appears throughout this manual, is the AI analysis feature built into OPENMARU APM. The AI buttons across the screens -- AI Diagnosis in the header (the whole group), the AI icon on a chart card (one chart), AI Diagnosis on an event, SQL AI Query Diagnosis, and CogentAI Analysis on a thread dump -- all lead to CogentAI. What is where is gathered in H14. Getting AI Analysis of Charts and Events, and the tutorial is in T3. Analysing a Cause with AI.

  • It works in a closed network too -- CogentAI runs on a small LLM of about the size of gpt-oss-20b, so it can be used in an environment with no Internet (a closed network) without calling an external AI service.

  • It needs a dedicated GPU server -- instead of using an external cloud AI, CogentAI runs on a GPU server separate from APM. The following components go on that server together.

    • vLLM -- the inference server that runs the LLM (the small model) quickly. It uses the GPU.
    • CogentAI Server -- the main body that receives AI analysis requests and orchestrates the LLM and tool calls.
    • Guardrail -- the safety mechanism that filters the questions and answers passing to and from the AI so they do not contain inappropriate or dangerous content.
    • RAG (Retrieval-Augmented Generation) + VectorDB (vector database) -- the part that searches by meaning for related material and past history and adds it as grounds for the answer.
    • Crawler -- the collector that gathers reference material such as manuals and technical documents and accumulates it in the VectorDB. The material RAG searches is created here.
    • MCP (Model Context Protocol) tools -- the standard interface connecting the AI so it can query real data directly. MCPs are attached per purpose -- APM, Database, Forecast, Chart, and others.

    In other words, using the AI features requires this dedicated AI GPU server (vLLM, CogentAI Server, Guardrail, RAG, VectorDB, Crawler, MCP) to be in place. The APM server alone does not run it.

    That said, because CogentAI uses a small LLM of about the size of gpt-oss-20b, a large GPU cluster is not needed -- one small server with a single GPU is enough to run all the components above.

  • It is optional -- the AI buttons appear only when an administrator has turned it on (Settings ▸ Cogent AI).

Note If you would like to use CogentAI, contact the Openmaru sales team -- sales@openmaru.io · 02-469-5426