Switch from OpenAI to Local Ollama

Swap your embedding and LLM to run fully offline with no API costs. Only config.yaml changes — your Python code stays the same.

embeddings:
  provider: "ollama"
  model: "nomic-embed-text"
  base_url: "http://localhost:11434"

llm:
  provider: "ollama"
  model: "qwen3.5:9b"
  base_url: "http://localhost:11434"
  num_ctx: 16384

Switch to OpenRouter (one key, many models)

OpenRouter gives you a single API key for both the LLM and embeddings, with free-tier models available. Requires pip install "ragwire[openrouter]" (Python ≥ 3.10).

embeddings:
  provider: "openrouter"
  model: "nvidia/llama-nemotron-embed-vl-1b-v2:free"
  api_key: "${OPENROUTER_API_KEY}"

llm:
  provider: "openrouter"
  model: "poolside/laguna-m.1:free"
  api_key: "${OPENROUTER_API_KEY}"

See the OpenRouter provider guide for details.

Recreate the collection when switching embedding models

Different embedding models produce different vector dimensions. If you already have a collection, set force_recreate: true once, run ingestion, then set it back to false.

For all available providers and their config values, see the Providers section.