API reference

Auto-generated from the live docstrings in src/verifiable_rag/. Use the sidebar to jump to a module, or browse the highlights:

Top-level entry points

• Pipeline — the core orchestration class
• ask() — single-shot one-liner for the simple case
• Presets — hybrid_balanced, local_minimal, etc.

Data model

• Models — Document, Sentence, Chunk, Answer, CitedSentence, VerificationResult, Span

Components (alphabetical)

• Chunkers — ParentChildChunker, ContextualChunker, the Chunker protocol
• Config — YAML loader, registry, register() extension hook
• Demo — bundled sample document helpers
• Embedders — BGE / Cohere / Voyage
• Generators — PromptedCitedGenerator, ConstrainedCitedGenerator, SAFECitedGenerator
• Indexers — HybridIndex, LanceDBIndex, BM25Index
• Parsers — DoclingParser, PyMuPDFParser, composite + caching wrappers
• Rerankers — Cohere / BGE
• Report — HTML audit report renderer
• Verifiers — DualNLIVerifier, HHEMVerifier, MiniCheckVerifier, LLMJudgeVerifier

Protocols

Every component is defined by a Protocol — swap implementations freely:

Protocol	Defined in	What it does
Parser	verifiable_rag.parsers	PDF → Document
Chunker	verifiable_rag.chunkers	Document → list[Chunk]
Embedder	verifiable_rag.embedders	list[str] → list[list[float]]
Reranker	verifiable_rag.rerankers	Rerank list[RetrievedChunk]
Generator	verifiable_rag.generators	Query + chunks → list[CitedSentence]
Verifier	verifiable_rag.verifiers	list[CitedSentence] + Document → list[VerificationResult]
NLIScorer	verifiable_rag.verifiers	list[(premise, hypothesis)] → list[float]