Tessera DB
PX-01
Views 04 and 05
Find a record by its words, or by what it is like.
BM25 full-text search and HNSW vector search are indexes over the same records, kept as they are written. Run them together and the results are fused with Reciprocal Rank Fusion.
01
03
Views 04 and 05 · Text and vectors
Search indexes that cannot fall behind the data.
A search engine fed by a sync job answers from the last sync. Here the indexes are kept on the write path, so a record is findable the moment it is written, and a vector search can find a row by its own text.
BM25
Full-text
HNSW
Vectors, SQ8
RRF
Hybrid fusion
0
Setup to start
02
03
What it does.
01
Full-text
BM25 over text properties, for names, codes and exact phrases.
02
Vectors
HNSW with SQ8 quantization. Send your own embedding, or plain text that the server embeds.
03
Hybrid
POST /api/search/hybrid runs both and fuses them, so records strong in either rise to the top.
04
Built-in embedder
Works offline with no setup. It matches spelling, not meaning: database and databases score 0.95, car and automobile 0.00. Set TESSERA_EMBEDDING_URL for a semantic model.
05
Scoped answers
Results are limited to the caller's tenant, and for end users to the rows their policies allow.
03
03
Both searches, from the kit.
# Keyword (BM25)
curl -s -H "Authorization: Bearer $TESSERA_KEY" "$TESSERA_URL/api/search?q=invoice&limit=5"
# Meaning: plain text in, nearest records out
curl -s -X POST -H "Authorization: Bearer $TESSERA_KEY" -H 'Content-Type: application/json' \
"$TESSERA_URL/api/search/vector-text" -d '{"text":"loan repayment","top_k":3}'