misakanet-core

Zero-dependency BM25 search engine with RRF fusion — extracted from MisakaNet.

• Pure Python, stdlib only
• BM25 ranking with configurable k1/b
• Metadata-weighted scoring
• RRF (Reciprocal Rank Fusion) for multi-query fusion
• CJK-aware tokenization
• Works in air-gapped environments

Installation

pip install misakanet-core

Usage

from misakanet_core import BM25, ScoredDocument, tokenize, rrf

# Prepare corpus
docs = [
    ScoredDocument("doc1", tokenize("the cat sat on the mat")),
    ScoredDocument("doc2", tokenize("the dog sat on the log")),
    ScoredDocument("doc3", tokenize("cats and dogs are friends")),
]

# Build index and search
engine = BM25(docs)
results = engine.search("cat dog", top_k=5)

for result in results:
    print(f"{result.doc_id}: {result.score:.4f}")

# Multi-query fusion with RRF
from misakanet_core import SearchResult, rrf
query1 = engine.search("cat")
query2 = engine.search("dog")
fused = rrf([query1, query2], top_k=3)

Why not use elasticsearch / tantivy / whoosh?

	misakanet-core	elasticsearch	tantivy	whoosh
Dependencies	Zero	JVM	Rust toolchain	Pure Python
Install time	0.5s	5min+	2min+	2s
Air-gapped	✅	❌	❌	✅
CJK support	✅	✅	⚠️	⚠️

License

MIT