Structured data

A pure RAG search finds passages of text, but it cannot calculate sums, quantities or rankings. For recurring document types – invoices, contracts, inspection reports – companyRAG can therefore additionally extract structured data from your documents: a language model fills a table schema you define, which is then queryable via SQL.

This gives you two access paths to the same collection, and they can be combined: semantic search across the text, and structured queries across the extracted fields.

Enable the Extract structured data (BETA) option when creating the collection.

Note

This feature is in beta. For critical evaluations, spot-check the extracted values against the original documents.

Defining the schema

The simplest case is one table with one row per document. For each column you define a name and a data type, and optionally mark whether the field is required or serves as a key.

Available data types: text, integer, numeric, boolean, timestamp, date.

Names must consist of lowercase letters, digits and underscores and start with a letter. Every row automatically receives the fields _id (UUID) and _ingested_at (timestamp).

You can also describe which information should be extracted, and limit the number of pages evaluated per document – for example to the first five pages, when the relevant information is always at the beginning. This reduces cost and runtime.

Multiple tables and relationships

For 1:n relationships – such as an invoice with several line items – add a second table, mark its key column there, and connect a column of the first table to it.

The connection is used as a join hint for queries. One document then produces several rows in the line-item table, while the header data is stored once in the first table.

Note

Deduplication across documents is based on exact matches: different spellings of the same value produce separate rows. Foreign key values are stored exactly as extracted – there is no automatic resolution.

Querying

The extracted tables are queried via the SQL tools of the ai-search MCP server, just like dataset collections. Which tools an agent needs for this is described in Using companyRAG in CompanyGPT.