Ultimate Local RAG-Pipeline

The problem

Building a RAG pipeline from scratch involves manually configuring vector databases, embedding models, and orchestration flows. Developers often waste hours debugging local Docker networking or writing complex SQL for hybrid search instead of building features.

What it does

• Deploys a local Postgres environment with PGVector and pre-configured schemas.
• Provides ready-to-import n8n workflows for document indexing and query retrieval.
• Integrates Ollama for local embedding generation and LLM inference.
• Sets up hybrid search functions to combine vector similarity with keyword matching.
• Automates the connection between n8n, Supabase, and local AI models.

Frameworks & tools

Supabase, PGVector, n8n, Ollama, Docker, and SQL.

Why this beats prompting it yourself

Getting local Docker services like n8n and Supabase to communicate with local Ollama instances requires specific networking configurations and SQL search functions that general LLMs often hallucinate. This kit provides tested JSON workflow exports and localized SQL schemas that work out of the box.

Use cases

• Building a private local knowledge base for proprietary documentation.
• Prototyping RAG applications without incurring OpenAI or Pinecone API costs.
• Processing sensitive data that must remain on-premises for GDPR compliance.
• Testing different embedding models and retrieval strategies locally.

Known limitations

Requires Docker and sufficient local RAM to run LLMs via Ollama. Performance depends on host hardware rather than cloud scaling.