Quickstart

The fastest path from a running instance to your first answer over your own document. It should take a few minutes once the platform is up.

Before you start

You need a running Minder instance (Self-hosting) and a reachable model — the default local Ollama, or an external one (AI setup). If a query later returns "Error generating response", the model backend is unreachable; fix that first.

There are two ways to do this: the control-plane UI (recommended the first time) or the HTTP API (for scripting). Both do the same four things — create a knowledge base, upload a document, build a pipeline over it, and ask a question.

Path A — the control-plane UI

1. Open the console and sign in. Browse to the client (over localhost a local account works; SSO needs real DNS/TLS — see Authentication). Most pages are browsable without login, but creating things needs you to be signed in.
2. Create a knowledge base. Go to RAG → Knowledge Bases, create one (give it a name and an optional description) — this is the data your questions are answered from.
3. Upload a document. Open the knowledge base and upload a file — PDF, Word, a spreadsheet, an image to OCR, a recording to transcribe, and many more formats. Upload runs as a background job: it comes back as processing and the page polls until it's completed (with chunk/vector counts) or failed. You can expand a document to inspect its stored chunks — useful for telling a bad extraction apart from a retrieval problem.
4. Create a pipeline. Go to RAG → Pipelines and create a pipeline over your knowledge base. A pipeline is what you actually ask questions against.
5. Ask a question. Open Ask, pick your pipeline, and ask. Conversations persist, so you can keep a thread going.

See Using Minder for the full tour of the console.

Path B — the HTTP API

All calls go through the API Gateway (http://localhost:8000 by default). See the API reference for the full surface and Authentication for obtaining a token.

# 1. Create a knowledge base (name required, description optional)
KB=$(curl -s -X POST http://localhost:8000/v1/rag/knowledge-bases \
  -H 'Content-Type: application/json' \
  -d '{"name":"My Docs","description":"my documents"}' | jq -r '.id')

# 2. Upload a document into it
curl -X POST "http://localhost:8000/v1/rag/knowledge-bases/$KB/upload" \
  -F "file=@report.pdf"

# 3. Create a pipeline over the knowledge base
PIPE=$(curl -s -X POST http://localhost:8000/v1/rag/pipeline \
  -H 'Content-Type: application/json' \
  -d "{\"name\":\"my-pipe\",\"knowledge_base_ids\":[\"$KB\"]}" | jq -r '.pipeline_id')

# 4. Ask a question
curl -s -X POST "http://localhost:8000/v1/rag/pipeline/$PIPE/query" \
  -H 'Content-Type: application/json' \
  -d '{"question":"What is in my docs?","top_k":3}' | jq '.answer'

From the command line

The open-source minder CLI wraps the same gateway: minder rag create-kb "My Docs" "my documents", minder rag pipelines, minder rag query <pipeline_id> "what is X?".

What next

• Tune retrieval — the query above used the default (dense) method. Minder also has HyDE, Self-RAG, auto, corrective, and RAPTOR methods, plus re-ranking, hybrid and parent-child retrieval. See RAG methods.
• Explore the knowledge graph — a second retrieval paradigm over the same documents (entities and relationships). See Using Minder.
• Extend the platform — add data sources and AI tools with plugins.
• Turn capabilities on and off — see Bundles.