RAG Versioning

Install

pip install briefcase-ai[rag]

Quick Example

from briefcase.rag import VersionedEmbeddingPipeline

pipeline = VersionedEmbeddingPipeline(
    model="text-embedding-3-small",
    version="v2",
)
embeddings = pipeline.embed(["document chunk 1", "document chunk 2"])

Architecture

RAG versioning tracks which embedding model, chunking strategy, and retrieval parameters were used for each decision. When you update your pipeline, previous decisions remain reproducible.

flowchart TD
    A["Documents"] --> B["Chunking"]
    B --> C["VersionedEmbeddingPipeline"]
    C --> D["Embeddings + version metadata"]
    D --> E["Vector store"]
    D --> F["DecisionSnapshot"]
    F --> G["Reproduce with same model + version"]

Key Classes

VersionedEmbeddingPipeline — wraps embedding calls with version metadata

Configuration

Option	Default	Description
`version`	`"v1"`	Pipeline version tag
`track_chunks`	`True`	Record individual chunk metadata