- Tech Lead · AWS OpenSearch
cat README.md
Open-source
full text & vector search on Parquet.
Millisecond queries across a billion docs, for 10× less.
infino bench --latency
1M docs · 384 dim · top-10
| percentile | latency |
|---|---|
| p50 | 3.75 ms |
| p90 | 6.10 ms |
| p99 | 9.40 ms |
| percentile | latency |
|---|---|
| p50 | 2.42 ms |
| p90 | 4.05 ms |
| p99 | 6.20 ms |
| percentile | latency |
|---|---|
| p50 | 3.66 ms |
| p90 | 5.95 ms |
| p99 | 9.10 ms |
infino cost
infino $1,631/mo
elasticsearch $22,155/mo 13.6×
opensearch $7,394/mo 4.5×
infino query "who builds infino"
02 Vinay Kakade
- Head of Eng · AWS OpenSearch
- Creator, ML platform · Lyft
03 Asif Makhani
- Sr. Dir, Search · LinkedIn
- GM, Image Search · Google
- Creator & Chief Architect · AWS OpenSearch
infino spec
- search
- full-text · vector · hybrid
- index
- BM25 (PFOR-delta, FST) · IVF + RaBitQ
- engine
- Infino
- language
- SQL (Apache DataFusion)
- storage
- object storage, S3 · GCS · Azure Blob · on-prem
- format
- Apache Parquet
- deploy
- on-prem · self-hosted cloud · hosted cloud
- license
- Apache-2.0
cat ARCHITECTURE.md
snowflakedatabricksbigqueryduckdbsparktrino infino
your data lake · s3 · gcs · azure blob · on-prem
your_data.parquet, one copy
row groups fts index vector index
↑ your data, unchanged ↑ embedded by infino, in place
cat query.sql
-- one query: SQL filter + full-text + vector, over Parquet SELECT id, title, snippet(body) FROM logs -- a Parquet dataset on S3 WHERE level = 'error' -- pushed-down filter AND text_match(body, 'disk full') -- BM25 full-text ORDER BY similarity(embedding, :q) -- vector ANN LIMIT 10; -- → ranked hits · one pass · no second store, no merge
infino init
→ index created on s3://your-bucket · apache-2.0