Industrial Maintenance Management Tools

AI-first maintenance management guide and training platform for industrial maintenance teams. The app uses a dark, minimal Next.js interface, Appwrite as the knowledge store, and MiniMax as the agent LLM.

What It Does

• Lets users ask maintenance management questions in Turkish.
• Runs a selectable multi-agent workflow:
  • CORE: strategy, BCM, RCM, criticality and reliability decisions
  • FIELD: preventive, predictive and autonomous maintenance procedures
  • FLOW: work orders, scheduling, backlog and resource flow
  • BASE: equipment hierarchy, minifiles and documentation structure
  • KPI: OEE, MTBF, MTTR and Wrench Time analysis
• Reads reference PDF/EPUB material from Appwrite collections.
• Skips agents that are not relevant to the selected question.
• Avoids showing source/citation details in user-facing answers.
• Converts [Diyagram Onerisi: ...] tags into diagram suggestion cards.

Stack

• Next.js App Router
• React
• Tailwind CSS
• Appwrite database
• MiniMax chat completions API
• Tavily web search fallback
• PDF/EPUB indexing scripts

Project Structure

app/
  api/ask/route.ts          API route for the agent workflow
components/
  maintenance-console.tsx   Main application screen
  agent-node.tsx            Agent status/timeline node
  agent-response-card.tsx   Agent answer card
lib/
  agents/                   Agent profiles, orchestration and MiniMax adapter
  appwrite/                 Appwrite server client and repositories
  knowledge/                Local fallback reference corpus and retrieval
  models/                   Shared TypeScript domain models
scripts/
  setup-appwrite.mjs        Creates Appwrite database collections/attributes
  index-reference-docs.mjs  Indexes PDF/EPUB files into Appwrite
public/reference-docs/      Local source folder for reference files
database/
  appwrite-collections.md   Appwrite collection design

Environment

Create .env.local in the project root:

MINIMAX_API_KEY=
MINIMAX_API_URL=https://api.minimaxi.chat/v1/chat/completions
MINIMAX_MODEL=MiniMax-M3

TAVILY_API_KEY=
TAVILY_API_URL=https://api.tavily.com/search

APPWRITE_ENDPOINT=
APPWRITE_PROJECT_ID=
APPWRITE_API_KEY=
APPWRITE_DATABASE_ID=bakim_rehber

APPWRITE_REFERENCE_DOCUMENTS_COLLECTION_ID=reference_documents
APPWRITE_REFERENCE_CHUNKS_COLLECTION_ID=reference_chunks
APPWRITE_EQUIPMENT_NODES_COLLECTION_ID=equipment_nodes
APPWRITE_WORK_ORDER_TEMPLATES_COLLECTION_ID=work_order_templates
APPWRITE_KPI_DEFINITIONS_COLLECTION_ID=kpi_definitions

Do not commit .env.local.

Setup

npm install
npm run appwrite:setup
npm run appwrite:index
npm run dev

Open:

http://localhost:3000

Reference Documents

Place PDF/EPUB files in:

public/reference-docs/

Then run:

npm run appwrite:index

The indexer:

• stores metadata in reference_documents
• stores text chunks in reference_chunks
• skips already indexed files by checksum
• skips unsupported files such as .xls and .xlsx

Appwrite Setup

Run:

npm run appwrite:setup

This creates the Appwrite database and collections if they do not already exist. It is intended to be idempotent.

Validation

npm run typecheck
npm run lint
npm run build

Notes

• The app is designed as an educational guide and decision-support simulator.
• It should not produce live operational decisions without reviewed source material.
• Agent answers are constrained by indexed reference documents.
• If no relevant local evidence exists and Tavily is configured, the agent can use web fallback evidence.
• Web fallback answers are marked in the UI, but source details are not shown in the answer text.