Skip to content

Part 0 - Prerequisites

Series: Tutorial index · You are here: Part 0 · Part 1 →

Before building the Document Q&A Assistant you need three backing services running locally and the Ollama model server installed with the models the tutorial uses.

What you need:

  • Docker and Docker Compose (Docker Desktop or Docker Engine + Compose plugin)
  • Python 3.11 or newer
  • Ollama - local model server

This takes about 10–15 minutes the first time (mostly waiting for model downloads).

1. Start the backing services

The tutorial uses PostgreSQL (database), Valkey/Redis (auth tokens + rate limiting), and Qdrant (vector store for the RAG pipeline).

Copy docker-compose.yml into the directory you'll use for the tutorial (e.g. ~/docqa):

cp /path/to/docs/tutorial/docker-compose.yml .

Or download it directly if you're working from the published docs:

curl -O https://raw.githubusercontent.com/your-org/fast-agent-stack/main/docs/tutorial/docker-compose.yml

Start all three services:

docker compose up -d

Verify they are healthy (all three should show healthy or running):

docker compose ps

You can also ping each service directly:

# PostgreSQL
docker compose exec postgres pg_isready -U docqa -d docqa

# Valkey/Redis
docker compose exec redis valkey-cli ping

# Qdrant
curl -s http://localhost:6333/healthz

2. Install Ollama

Download and install Ollama from ollama.com for your OS (macOS, Linux, or Windows). After installation, the ollama CLI is available in your terminal.

Confirm Ollama is running:

curl http://localhost:11434/api/tags

You should get a JSON response. If Ollama isn't running, start it:

ollama serve

3. Pull the tutorial models

The Document Q&A Assistant uses three Ollama models across different tutorial parts:

# Chat model - used from Part 1 onwards for the conversational agent
ollama pull llama3.2

# Embedding model - used in Part 4 to embed uploaded documents
ollama pull nomic-embed-text

# Reranking model - used in Part 5 to rerank retrieved document chunks
ollama pull qllama/bge-reranker-v2-m3

Disk space: llama3.2 is ~2 GB, nomic-embed-text is ~274 MB, and qllama/bge-reranker-v2-m3 is ~570 MB. Make sure you have at least 4 GB free.

Verify the pulls succeeded:

ollama list

You should see all three models listed with their sizes.

4. Prepare your environment file

The tutorial app reads connection details from a .env file. You'll create this in Part 1 when you scaffold the project, but knowing the values now is useful.

The connection strings for the services above are (note the DOCQA_ prefix - the framework namespaces all env vars by project name):

DOCQA_DATABASE_URL=postgresql+asyncpg://docqa:docqa@localhost:5432/docqa
DOCQA_REDIS_URL=redis://localhost:6379
DOCQA_QDRANT_URL=http://localhost:6333

# Ollama for LLM and embedding (OpenAI-compatible endpoint)
DOCQA_LLM_BASE_URL=http://localhost:11434/v1
DOCQA_LLM_MODEL=llama3.2
DOCQA_LLM_API_KEY=ollama
DOCQA_EMBEDDING_PROVIDER=openai
DOCQA_EMBEDDING_BASE_URL=http://localhost:11434/v1
DOCQA_EMBEDDING_MODEL=nomic-embed-text
DOCQA_EMBEDDING_API_KEY=ollama

You do not need to create this file yet - fas new generates a .env.example you'll copy and fill in during Part 1.

Checklist

Before continuing, confirm:

  • docker compose ps shows postgres, redis, and qdrant as healthy
  • curl http://localhost:11434/api/tags returns JSON
  • ollama list shows llama3.2, nomic-embed-text, and qllama/bge-reranker-v2-m3

Next steps

You're ready to build. Head to Part 1 - Scaffold to scaffold the docqa project.