Infino
sign up talk to an engineer
# no parquet yet? start from ingest →

Integrates with
your data stack.

Point Infino at Parquet you already have, or append JSON directly. It builds the BM25 index and writes hydrated Parquet back to your bucket. Embeddings can come from the table’s ingest model, from vectors you send, or from a later hydration pass.

Sign up How it works
your app REST · SQL · Query DSL warehouse & lakehouse snowflake · databricks bigquery · duckdb · spark OBJECT STORAGE docs.sf.parquet logs.sf.parquet

infino hydrate s3://your-bucket/logs # the files stay where they are

How you adopt it

Hydration reads the files where they already sit and writes them back to the same prefix. If you have no Parquet yet, append JSON and Infino writes it for you.

  1. 01 Existing Parquet Already on S3, GCS, or Azure Blob
  2. 02 Hydration BM25 index; embeddings generated or supplied
  3. 03 Hydrated Parquet Same file, now searchable
docs.json batch append infino docs.sf.parquet bm25 · vectors

Starting without Parquet →

cat query.sql # search and sql, one pass

What one statement expresses over that file

Retrieval, the filters, and the ranking are one statement.

query.sql 
SELECT   _id, title, score
FROM     hybrid_search(                       -- BM25 + vector, fused by RRF
           'logs', 'body', 'disk full',       --   the text side
           'embedding', :q, 50                --   the vector side, top 50
         )
WHERE    level = 'error'                      -- pushed-down filter
  AND    ts > now() - interval '24 hours'
ORDER BY score DESC
LIMIT    10;

-- → ranked hits · one pass over Parquet

RAG: joins, group-bys, and windows over the same result set →

infino diff my-stack # what each tool does today

What it replaces, and what it runs beside

replaces

tool what it does today with infino
elasticsearch · opensearch BM25 and vector search over engine-private segment files on cluster disks, sized and paid for at peak. The same ranking over Parquet in your own bucket.
vector database A second copy of the embeddings in a second store, with keyword and vector hits merged in your client code. One index. BM25 and vector run in one pass and are fused by RRF inside the query.

runs beside

Infino does not replace a warehouse. Teams move query workloads onto Infino to reduce cost and improve performance, but Infino does not have the ecosystem of a typical data warehouse.

tool what it does today with infino
snowflake · databricks · bigquery SQL and aggregates over the lake. Ranked retrieval means a table scan, or a separate managed search service with its own copy. Move retrieval and selected SQL query workloads onto Infino. The warehouse keeps BI and governance; both read the same Parquet files.
duckdb · spark · trino They already read your Parquet, and there is no BM25 or ANN index inside the file for them to use. Hydration puts that index in the file. They keep reading the columns; Infino serves the ranked read.

The full comparison →

cat catalog.md

Iceberg, Delta Lake, and Hudi

Infino hydrates the data files your table already points at, then registers the swap with the catalog. The catalog stays the ACID authority; a query reads one pinned snapshot.

registers the swap catalog iceberg · delta · hudi manifest *.parquet hydrates infino

infino spec

search
full-text · vector · hybrid
index
BM25 (PFOR-delta, FST) · HNSW · OPANN + Sq16
engine
Rust
language
SQL (Apache DataFusion) · REST · Query DSL
storage
object storage, S3 · GCS · Azure Blob · on-prem
format
Apache Parquet
deploy
on-prem · self-hosted cloud · hosted cloud
core license
Apache-2.0

Go deeper: how it works · elasticsearch replacement · agent observability · RAG · docs ↗

Sign up Talk to an engineer