Launching today
Infino AI
Search, vectors and SQL over one Parquet copy, for agents
35 followers
Search, vectors and SQL over one Parquet copy, for agents
35 followers
Infino is an open-source retrieval engine for AI agents. It keeps one copy of your data as Parquet on S3 or local disk, with BM25 and vector indexes inside the files. Agents run keyword, vector, hybrid and SQL search on it. Apache-2.0. Infino Cloud is generally available, with a free tier.
Free Options
Launch Team
Atlassian LoomTalk it through in Loom. Ship with AI
Promoted
Hi Product Hunt 👋 We're the team behind Infino. We built search and ML systems at Google, LinkedIn and Amazon, including AWS OpenSearch, and we just raised $7.5M from Bessemer to build one retrieval layer for agents.
Agents don't query like people. One question becomes dozens of keyword lookups, vector searches, filters and joins, and today each of those lives in a different system with its own copy of the data.
Infino keeps one copy. Your data stays as Parquet on S3 (or local disk), with the BM25 and vector indexes stored inside the same files. Any Parquet reader can still open them. On top, one engine runs keyword, vector and hybrid search, and search results are SQL tables you can join and aggregate.
Try it in 30 seconds:
No install: watch Claude Code answer the same question on Infino, Elastic and Snowflake side by side in our live demo, or point it at your own files
Local: pip install infino or npm install @infino-ai/infino
For agents: the Infino MCP server, so Claude Code or Cursor can search your data directly
Hosted: Infino Cloud, free to start
One number we're proud of: filtered to the search engines on ClickBench, Infino ranks #1 for SQL analytics (combined metric), ahead of Elasticsearch, Quickwit, ParadeDB and SigLens. See the board
The engine is Apache-2.0 on GitHub. We'd love your hardest retrieval questions, and we're here all day to answer them.
one parquet copy for sql and retrieval is an appealing place to start
less data syncing to look after would be a big part of the value for a small team
Search and SQL over the same Parquet files is a neat approach. Congrats, team!
Being able to join search results with the original rows is the detail that stands out. That's useful when you're trying to understand why a hybrid search returned a particular match.