ragpick¶
Choose, build, evaluate, and run the right RAG architecture.
Most RAG projects don't fail because of the LLM. They fail because the architecture was picked by guesswork: vector search where exact product codes matter, no access control for internal documents, no evidence that the "improved" pipeline is actually better.
ragpick turns that choice into an engineering process:
- Describe your requirements (corpus size, keyword-heavy queries, latency, access
control) and
ragpick recommendexplains which blueprint fits and why. - Define the pipeline as YAML. Architecture as code, validated with readable errors.
- Evaluate competing blueprints on your data with
ragpick evaland pick the winner from numbers, not vibes.
ragpick sits next to LangChain and LlamaIndex rather than replacing them. It owns the interfaces, config, selector and evaluation; other libraries plug in as components.
Install¶
pip install ragpick # core, offline components
pip install "ragpick[openai]" # + OpenAI embeddings and chat models
pip install "ragpick[all]" # + sentence-transformers, Qdrant, PDF
ragpick 1.0 is stable: see the stability policy and the upgrade guide.
Next steps¶
- Concepts: components, blueprints, the selector, access control.
- Blueprints: what each architecture is good at.
- Config reference: every
rag.yamlkey. - Writing a plugin: add your own loader, store, or model.
- Evaluation: metrics, generated datasets, CI gates and
bench. - Routing, serving, multi-tenancy, conversations and feedback: running it in production.
- Tracing & caching and LangChain & LlamaIndex.
- Stability and upgrading: the 1.x contract.
- Design decisions: the reasoning behind every release.