Getting the Chatbot to Answer from Our Documents
This tutorial walks you through confirming that answers are actually grounded in our company's documents.
From step 2~3 onward, where you add and index a document, you'll need the installed stack.
- A completed AI stack installation → One-line Install
- An admin console login account → Changing the Initial Password
- One document file to test with (equipment manual, work standard, etc.)
There's a public demo — https://demo-ai.plantpulse.io (admin / admin123!).
⚠️ Only the Copilot is open in the demo. The admin console (GPU, LLM, RAG screens) is not included —
see the AI Admin Console docs for that.
Step 1 — Check in the admin console that everything is alive
First, log in to the Admin Console. In the summary line at the top of the home screen,
confirm that 중단 is 0.
정상 12 주의 0 중단 0 합계 12
If there's a red light, stop here and fix that service first — every later step builds on top of it.
If XID under GPU on the rail is non-zero, that's a hardware/driver error. Even if things work now, they'll stop soon → Admin Console.
Step 2 — Add one document
On the rail, go to RAG API → Documents. Upload a document.
First checkpoint — the document should appear in the list and finish indexing. Also check Sync History to confirm the last run finished successfully.
A deleted document can no longer be used by the chatbot as grounding. If you don't have the original file on hand, don't delete it.
Step 3 — Test directly whether search works
In RAG API → Query Test, search using wording a user might actually ask.
If content from the document you just added shows up in the results, you've passed the second checkpoint.
If it doesn't show up, try changing the search method — you can choose among
hybrid (default) · local · global · naive
mix. If it still doesn't show up, go back to step 2 and check whether indexing actually finished.
Step 4 — Ask the chatbot
Opening the Copilot takes you to the Ask home screen — it has a greeting and an input box, and if there's something that needs your attention right now, a collapsed line reading "N tasks for today" appears below it (click to expand it; if there's nothing to do, this line doesn't appear) → Copilot Screen Map
Ask the same question as in step 3 in the input box. If the answer comes back grounded in the document you just added, you're done → AI Chat Web.
Check whether the document you just added appears in the grounding card below the answer. Clicking the card opens that document's original text in a panel on the right, next to the conversation (for formats other than PDF, use Download Original in the panel).
If search works fine in the console but the chatbot's answer seems off, the problem is on the LLM side, not search. Check the response speed (P95) and status under LLM on the rail.
Step 5 — Ask about equipment data too
Besides documents, you can also ask about real-time and historical data. This path goes through the platform's integrated MCP → MCP — Real-time Data Gateway.
"What was the uptime for Line 3 yesterday?"
Once you're done
| What to do next | Document |
|---|---|
| I want to approve anomalies/tasks found by AI and see them through to completion | Proposals |
| I want to delegate checks I do every day | Automation |
| I want to understand how anomalies are caught and turned into incidents | Anomaly Detection · Incident Response |
| I want to see what kinds of things I can ask | Example Questions |
| Operational inspection habits | AI Admin Console |