Emmimal / temporal-rag

What is temporal-rag?

Temporal-rag is a Python library that adds a post-retrieval temporal layer to any RAG pipeline, fixing how standard vector search blindly retrieves stale or expired documents. After your retriever pulls candidates, it runs validity filtering to hard-remove outdated info, classifies documents by kind (static facts, versioned updates, time-bound events), applies time decay and recency scoring, then hybrid reranks with semantic similarity. One `retrieve()` call handles it all, no heavy dependencies beyond numpy.

Why is it gaining traction?

It stands out on temporal rag github by slotting into existing RAG setups without reindexing or new infra—pure Python, 15-30ms latency on 20 docs. Demos contrast naive RAG failures (expired policies topping results) against temporal rag pipeline wins, like surfacing active events over old versions. Advanced patterns like adaptive weighting from query text and failure logging make production tweaks straightforward.

Who should use this?

RAG engineers maintaining docs, APIs, or policies where superseded facts cause bad answers, like support bots citing deprecated endpoints. Teams handling news, research, or announcements need its event classification and decay to prioritize fresh signals. Skip if your corpus is fully static.

Verdict

Grab it for temporal rag needs—MIT licensed, excellent docs, reproducible demos make evaluation instant despite 26 stars and 1.0% credibility score signaling early maturity. Solid starting point, tune configs per domain before prod.

(198 words)