What should I use to build a private Q&A app over our knowledge base?

For a private Q&A app over your knowledge base, the best default choice is usually a retrieval-augmented generation (RAG) stack.

Recommended approach

Use:

1. An LLM for answering questions
2. Embeddings to represent your documents
3. A vector database to store/search those embeddings
4. A document ingestion pipeline to keep the knowledge base up to date
5. Auth + access control to keep it private

Good stack options

If you want the fastest path

• OpenAI or Anthropic API for the model
• Postgres + pgvector for vector storage
• LangChain or LlamaIndex for orchestration
• Next.js / React for the app UI
• Auth0, Clerk, or Supabase Auth for login

This is a very practical setup for most teams.

If you want enterprise/private hosting

• Self-hosted model like Llama 3 or Mistral
• Postgres + pgvector, Qdrant, or Weaviate
• LlamaIndex for indexing/retrieval
• FastAPI / Node.js backend
• SSO integration with Okta/Azure AD/Google Workspace

If you want the simplest internal tool

• Supabase
  • auth
  • Postgres
  • pgvector
  • storage
• LlamaIndex
• OpenAI API
• Next.js frontend

This is often the easiest way to ship quickly.

---

What matters most

For a private KB Q&A app, prioritize:

• Permission-aware retrieval: only fetch documents the user is allowed to see
• Good chunking: split docs sensibly before embedding
• Citations/source links: users should see where answers came from
• Freshness: reindex when docs change
• Hybrid search: combine semantic search with keyword search if your KB is large or technical

---

My practical recommendation

If you’re building this now, I’d start with:

• Next.js
• Supabase
• pgvector
• LlamaIndex
• OpenAI GPT-4.1 or similar
• Postgres row-level security for privacy

That’s a strong balance of speed, quality, and maintainability.

If you want, I can also suggest:

• a specific architecture diagram
• a tech stack by budget
• or a minimal MVP plan for building it in 1–2 weeks.