RAG API Reference¶

Document¶

A document for retrieval-augmented generation.

from flowgentra_ai import Document

Document(id: str, text: str, metadata: dict | None = None)

A result from a vector store search.

A chunk of text from text splitting.

Property	Type	Description
`text`	`str`	Chunk text
`metadata`	`ChunkMetadata`	Chunk metadata (source, index, offsets)

Embedding generation from text.

from flowgentra_ai import Embeddings

Method	Description
`Embeddings.mock(dimension)`	Hash-based mock (no API needed)
`Embeddings.openai(api_key, model="text-embedding-3-small")`	OpenAI embeddings
`Embeddings.openai_with_dimension(api_key, model, dimension)`	OpenAI with custom dimension
`Embeddings.openai_cached(api_key, model="text-embedding-3-small")`	OpenAI with cache
`Embeddings.ollama(model, base_url=None, dimension=None)`	Ollama (local)
`Embeddings.mistral(api_key, model=None)`	Mistral embeddings
`Embeddings.huggingface(model, api_key, endpoint=None, dimension=None)`	HuggingFace

In-memory vector store with cosine similarity search.

from flowgentra_ai import InMemoryVectorStore

InMemoryVectorStore()

Method	Returns	Description
`index(doc, embedding)`	`None`	Index a document with its embedding
`search(query_embedding, top_k=5, filter=None)`	`list[SearchResult]`	Search by embedding
`get(doc_id)`	`Document`	Get document by ID
`delete(doc_id)`	`None`	Delete by ID
`list()`	`list[Document]`	List all documents
`clear()`	`None`	Remove all documents

Configuration for retrieval strategies.

from flowgentra_ai import RetrievalConfig

Method	Description
`RetrievalConfig.semantic(top_k=5, threshold=0.7)`	Semantic search config
`RetrievalConfig.hybrid(keyword_weight=0.3, top_k=5, threshold=0.7)`	Hybrid search config

Property	Type	Description
`top_k`	`int`	Number of results
`similarity_threshold`	`float`	Minimum score threshold

Full retrieval pipeline: embed, search, hybrid, rerank, dedup.

from flowgentra_ai import Retriever

Retriever(store: InMemoryVectorStore, embeddings: Embeddings, config: RetrievalConfig)

Method	Returns	Description
`retrieve(query)`	`list[SearchResult]`	Execute the full retrieval pipeline
`with_dedup(threshold)`	`None`	Enable deduplication

An extracted PDF document.

Function	Returns	Description
`chunk_text(text, chunk_size, overlap=0)`	`list[str]`	Split text by character count
`chunk_text_by_tokens(text, max_tokens, overlap_tokens=0)`	`list[str]`	Split by token count
`extract_text(path)`	`str`	Extract text from a PDF
`extract_pdf(path)`	`PdfDocument`	Extract PDF as object
`extract_and_chunk(path, chunk_size=1000, overlap=200)`	`list[(str, str)]`	Extract + chunk
`estimate_tokens(text)`	`int`	Estimate token count
`bm25_score(query, documents)`	`list[float]`	BM25 keyword scores
`hybrid_merge(results, query, keyword_weight=0.3)`	`list[SearchResult]`	Merge semantic + keyword
`dedup_by_id(results)`	`list[SearchResult]`	Deduplicate by ID
`dedup_by_similarity(results, threshold=0.85)`	`list[SearchResult]`	Deduplicate by similarity
`decompose_query(query, max_depth=2)`	`list[str]`	Split compound queries
`load_document(path)`	`LoadedDocument`	Load a file
`load_directory(path)`	`list[LoadedDocument]`	Load all files in a directory