OpenSearch veterans launch Infino. Here’s why it matters for agent builders.

Infino has formally launched an agent retrieval platform built around a simple bet: AI agents don’t query data like humans do, so the data stack should stop pretending they do.

CEO Ekechi Nwokah says agents are now the “largest new consumer of data since the web browser,” but they’re still being fed through systems built for older patterns: SQL warehouses, keyword search, vector search, and ETL pipelines that keep everything in sync. Infino’s pitch is to collapse that mess into one layer over object storage, with one copy of the data and one interface for ranked search, exact counts, joins, filters, aggregation, and reasoning.

That matters because agents are noisy readers. They ask lots of small questions, often in parallel, then stitch the answers together themselves. Each round trip eats context, time, and money. Infino argues that a warehouse here, a search engine there, and a vector database somewhere else just forces the agent to do extra work, and then do more work to verify the first round of work.

The company’s answer is to store data as Apache Parquet on object storage and put search indexes beside the Parquet footer. That lets other Parquet tools keep reading the same files, while Infino’s engine handles retrieval. The company says its open source core is Apache-2.0 on GitHub, and its hosted Infino Cloud can point at existing Parquet data without teams having to build infrastructure first.

Infino also says it has built targeted inference models into the retrieval loop, because agents spend a lot of time bouncing between query formulation, search, validation, and retries. On the business side, Nwokah claims the platform is about 10x cheaper than traditional search or analytics infrastructures for its intended workload, and its own published comparison pegs it at about 10.5x cheaper than Elasticsearch and 23x cheaper than OpenSearch.

The startup says it is already handling multi-billion-document use cases in people search, document processing, product analytics, and security, with one unnamed customer putting petabytes of company data on the system. Nwokah and his co-founders, who include veterans of Amazon, Google, LinkedIn, and AWS OpenSearch, are betting that the next big retrieval problem is not better search in isolation. It’s making agents stop juggling five tools just to answer one question.