the open source core (in every tier):
everything in github, plus:
everything in enterprise, plus:
The entire engine. The open-source core is Apache-2.0 with no feature gates: BM25, vector, and hybrid search, SQL through DataFusion, crash-safe commits, snapshots and time-travel, single-node, PB-scale, on your object storage. pip install infino, npm install @infino-ai/infino, or cargo add infino.
No. Infino is a retrieval engine you run in your own process, against S3, Azure Blob, or local disk, with no server or cluster to stand up. Cloud is for teams that want the same engine run for them, with support and SLAs.
Three things: data stored (logical GB per month), data written, and data queried. The calculator above uses the exact rates. The $16/mo minimum is a floor, not a base fee, usage isn’t billed on top of it.
Vectors are metered as f32, dimensions × 4 bytes per document, plus attribute bytes. Full-text-only workloads don’t pay for dimensions at all; the fts tab in the calculator reflects that.
Their architecture bills for peak, always: a cluster sized for the hottest hour, replicas that multiply the data, JVM heap headroom, block storage, snapshots, and cross-AZ traffic. Infino’s durable copy is your bucket, no replicas to fund, retention at object-storage rates, and the searchable copy measured about a third of the raw corpus at 10M docs.
After an indexing pass, yes. Ordinary Parquet doesn’t contain retrieval structures, so Infino embeds the BM25 and vector indexes into the files, in place, no copy into a proprietary store. One caveat: rewriting a file through an ordinary Parquet writer drops the indexes.
Same commercial engine. Enterprise you run yourself, multi-node, HA, RBAC, audit, multi-tenancy, on-prem or in your VPC. Cloud is that engine run by us: managed SaaS, production support, SLAs.
Apache-2.0. Permissive, no copyleft, no usage restrictions, the LICENSE file in the repo is the standard text, unmodified.
