Configuration

version: "1.0"
 
# Pipeline type
pipeline: naive   # naive | hybrid | self | long_context | agentic
 
plugins:
 
  # LLM - the model that generates answers
  llm:
    provider: anthropic           # anthropic | openai | mistral | groq | llama | local
    model: claude-sonnet-4-6
    api_key: ${ANTHROPIC_API_KEY} # or hardcode directly
    temperature: 0.2
    max_tokens: 1024
 
  # Embeddings - converts text to vectors
  embedding:
    provider: openai              # openai | bge | cohere | sentence_transformer
    model: text-embedding-3-small
    api_key: ${OPENAI_API_KEY}
    batch_size: 32
 
  # Vectorstore - stores and searches vectors
  vectorstore:
    provider: faiss               # faiss | chroma | pinecone | weaviate | qdrant | pgvector
    index_path: .ragway/index
 
  # Retrieval strategy
  retrieval:
    strategy: vector              # vector | bm25 | hybrid | multi_query | parent_document
    top_k: 5
    hybrid_alpha: 0.5             # hybrid only: 0=bm25, 1=vector
 
  # Reranker - improves result ordering
  reranker:
    enabled: true
    provider: cohere              # cohere | bge | cross_encoder
    model: rerank-english-v3.0
    api_key: ${COHERE_API_KEY}
    top_k: 3
 
  # Chunking - splits documents into pieces
  chunking:
    strategy: recursive           # fixed | recursive | semantic | sliding_window | hierarchical
    chunk_size: 512
    overlap: 50

  llm:
    provider: anthropic
    api_key: ${ANTHROPIC_API_KEY}  # reads from environment

export ANTHROPIC_API_KEY=sk-ant-...