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

Infino, a startup founded by former OpenSearch veterans, has launched an agent retrieval platform that aims to address the fragmented data query paradigm currently employed by AI agents. The company's platform promises to provide a single retrieval layer, allowing all AI agents to query all data from one place.

Infino CEO Ekechi Nwokah explained that agents have changed the data query paradigm and no longer query like humans. Traditional data query systems, such as data warehouses, search engines, vector databases, and ETL pipelines, were optimized for human or application queries, not the unique data reader dynamic of AI agents. Agents issue many small questions to answer one large query, often simultaneously, and each result they receive can add cost and latency to subsequent model calls.

Infino's solution involves storing data in Apache Parquet format on object storage and allowing agents to directly search, rank, filter, join, aggregate, and reason over the data. This approach consolidates the data store, enabling agents to perform complex queries in milliseconds using a single natural language English question. Infino's retrieval layer provides a unified SQL interface over object storage, combining structured and unstructured queries in a single query.

The company's approach eliminates the need for multiple systems, ingest paths, ETL jobs, schemas, and glue code. Developers and data teams can enforce centralized policies for agent data access, governance, and permissions. Infino's platform is designed to be open source, with a focus on efficient data storage using Apache Parquet files with embedded search indexes. This allows any tool that reads Parquet files to access the data both with and without Infino's additional features.