using-infino

Using Infino

Infino is a retrieval engine over data on object storage: full-text (BM25), vector, hybrid, and SQL search over one copy of the data, embedded in-process against a local path or bucket, or against a hosted Infino Cloud database. These tools (from the Infino MCP server) operate on a connected catalog of tables. The full toolset is always available; you are responsible for the data you change, and on Infino Cloud the API key's capabilities decide what the connection may do.

1. Discover before you search

• infino_list_tables: list the tables in the catalog. Start here when you don't already know the table name. On a hosted connection a 404 here means the database does not exist yet: call infino_create_database.
• infino_describe_table: a table's columns and types, so you know which column to target and what each result row carries. It does not report indexes; a search on a column with no index says so in its error.

2. Pick the search tool by the question shape

• infino_keyword_search: literal terms such as identifiers, error codes, product names, exact phrases. Ranked BM25. Use when the wording is known.
• infino_semantic_search: meaning or paraphrase when the exact wording is unknown. Pass filter ({column, query}) to first restrict to rows whose keyword column matches, then rank semantically within them.
• infino_hybrid_search: the query has both specific terms and an intent. Fuses keyword + vector in one ranking pass. A good default when unsure.
• infino_sql: structural or analytical questions: counts, GROUP BY, joins, aggregates, filtering by exact column value. The search functions are callable inside SQL too (bm25_search(...), and vector_search(...) with a {{name}} placeholder plus embed: {name: "text"}).
• infino_token_match / infino_exact_match / infino_count: unranked filters and a tally, when you need the set or the number, not an order.

Prefer the dedicated search tools over hand-written SQL for retrieval; they embed the query and project results for you.