Quickstart: pipeline

Run a retrieval-augmented generation pipeline end-to-end with RagPipeline.

from attune_rag import RagPipeline, DirectoryCorpus

pipeline = RagPipeline(corpus=DirectoryCorpus("docs/"))
answer, result = pipeline.run_and_generate(
    query="How do I configure a workflow?",
    provider="openai",
)
print(answer)
print(f"Confidence: {result.confidence:.2f}  |  Elapsed: {result.elapsed_ms:.0f} ms")

Expected output:

To configure a workflow, open the attune dashboard and ...
Confidence: 0.87  |  Elapsed: 312 ms

Prerequisites

Install the package: pip install attune-rag
Set the environment variable for your LLM provider (for example, OPENAI_API_KEY)

Steps

1. Build a corpus

Point DirectoryCorpus at a folder of documents. The pipeline retrieves grounding context from these files at query time.

from attune_rag import DirectoryCorpus

corpus = DirectoryCorpus("docs/")

2. Create the pipeline

Pass the corpus to RagPipeline. All other arguments — retriever, expander, reranker — are optional and default to built-in implementations.

from attune_rag import RagPipeline

pipeline = RagPipeline(corpus=corpus)

3. Run a query

Call run_and_generate with your query and LLM provider. It returns the generated answer string and a RagResult containing provenance metadata.

answer, result = pipeline.run_and_generate(
    query="How do I configure a workflow?",
    provider="openai",
)

print(answer)
print(result.citation)       # CitationRecord with source provenance
print(result.fallback_used)  # True if no grounding context was found

If result.fallback_used is True, the corpus contained no relevant documents for the query. The pipeline still responds, but without grounding context.

4. Format citations

Turn the citation provenance into readable Markdown for display or logging.

from attune_rag import format_citations_markdown

print(format_citations_markdown(result.citation))