Search or Chat returns no useful results

Problem

Search returns no matches, Chat gives vague answers, says it cannot find information, or shows configuration errors.

What it means

Search and Chat depend on a chain: indexed documents → vector retrieval → LLM generation. A break at any step produces poor or empty results. Not every question can be answered — only questions grounded in indexed content have a fair chance of a useful answer.

Diagnostic sequence

Work through these checks in order:

1. Documents available and Ready

• Open Knowledge Base → Documents list.
• Confirm at least one document shows Ready (not only Processing or Error).
• If nothing is Ready, finish processing or fix failed documents first.

2. Knowledge Base content matches the question

• Ask about topics actually present in uploaded documents.
• Try a query using exact terms from a known document title or section.

3. LLM connection

• Settings → LLM must show Connected.
• Run Test connection.
• See LLM connection problems.

4. Vector Database connection

• Settings → Vector DB must show Connected.
• Run Test connection.
• See Vector Database connection problems.

5. License and admin readiness

• Settings → License should show a valid active plan.
• If the admin shows Admin not ready, complete license, LLM, and Vector DB setup.

6. Query quality

• Use clear, specific questions.
• For Search, try shorter keyword-style queries.
• For Chat, one focused question often works better than a long multi-part prompt.

7. Activity log

• Open Settings → Activity log for NX_RETRIEVAL_FAILED, NX_LLM_CALL_FAILED, or NX_PROVIDER_CONFIG_INVALID.

How to resolve

Finding	Action
No Ready documents	Upload content and wait for Ready
Processing stuck	See document processing problems
LLM / Vector DB not connected	Fix provider settings and test connections
Retrieval errors in log	Check embedding key, Weaviate collection, schema, reindex if needed
Question off-topic	Rephrase or add relevant documents

Expected result

With Ready documents, connected providers, and a relevant query, Search returns matching snippets and Chat produces answers grounded in your Knowledge Base with sources when enabled.

What not to expect

Noulek does not answer from general world knowledge alone when RAG is configured — it retrieves from your indexed content first. A query with no relevant indexed material may correctly produce a limited or empty result.