privaterag-langchain 0.1.0

LangChain retriever adapter for the PrivateRAG Python SDK.

privaterag-langchain

Thin LangChain retriever adapter for the core PrivateRAG Python SDK.

The adapter keeps LangChain imports isolated to this package. The core privaterag SDK does not import LangChain.

Install

Published packages declare the core privaterag SDK as a dependency. If LangChain is not already present in the environment, install the optional extra:

python -m pip install privaterag-langchain[langchain]

The v0 supported framework range is langchain-core>=1.0,<1.4; see ../../docs/adapter_compatibility.md for the tested lower-bound and current-version matrix.

For local editable development:

python -m pip install -e python/privaterag
python -m pip install -e "python/privaterag-langchain[langchain]"

Usage

from privaterag import PrivateRAGClient
from privaterag_langchain import PrivateRAGRetriever

client = PrivateRAGClient.from_manifest(
    "http://127.0.0.1:18090/manifest.json",
    embed_query=local_embed_query,
)

retriever = PrivateRAGRetriever(
    client,
    top_k=2,
    retrieval_count=4,
)

documents = retriever.get_relevant_documents("What needs review before onboarding?")

top_k must be positive. retrieval_count, when provided, must be positive and greater than or equal to top_k; otherwise the adapter fails before delegating to the SDK. See ../../docs/padding_policy.md.

Returned LangChain Document objects use retrieved chunk text as page_content. Metadata includes the original chunk metadata plus chunk_id, document_id when present, score when present, and namespaced privaterag_* copies of those retrieval fields.

The adapter delegates retrieval to privaterag.PrivateRAGClient. It does not implement embedding providers, LLM calls, reranking, server-side vector search, or direct chunk lookup. The current supported simplepir-rs-v0-1000-1k backend is supported only within its documented v0 limits and remains unaudited, not externally audited, and not production-grade cryptographic assurance. The old simplepir-rs-experimental-v0 profile remains smoke-scale, experimental, unaudited, and compatibility/development-only.