knowledge

Knowledge Sources Example

Demonstrates RAG document ingestion with three source types working together:

• TextFileSource — unstructured text (DraftCrew overview)
• CSVSource — structured tabular data (feature tracker)
• JSONSource — hierarchical data (build metrics)

The agent indexes all three files via draftRAG (Ollama embedder) and answers questions that require knowledge from different sources.

Prerequisites

• Ollama running (default: http://localhost:11434)
• An embedding model: ollama pull nomic-embed-text or ollama pull embeddinggemma:300m
• An LLM model: ollama pull gemma3:4b

Usage

Ollama LLM + Ollama embedder (recommended)

OLLAMA_BASE_URL=http://localhost:11434 \
  OLLAMA_MODEL=gemma3:4b \
  LLM_EMBED_MODEL=embeddinggemma:300m \
  go run main.go

OpenAI LLM + Ollama embedder

OPENAI_API_KEY=sk-... \
  LLM_EMBED_MODEL=nomic-embed-text \
  go run main.go

Customising the embedder

What the agent answers

1. Which process types does DraftCrew support? (from txt)
2. Which features are done vs pending? (from csv)
3. Current version, packages count, test coverage? (from json)
4. What RAG engine does it use? (from txt + json)

Notes

• Sources are loaded eagerly at agent init — one shared Pipeline (embedder + store) for all sources
• Each source is idempotent: calling Load() twice is a no-op
• If a source file is missing or malformed, the agent logs a warning and continues without it