Data Tables

Data Tables turns collections of similar documents into a structured, persistent project table. Each file is kept as its own record and remains connected to its extracted text and business fields. Typed columns make results filterable and comparable.

A computed column defines which information should be produced from existing fields. Processing runs per record and distinguishes queued, running, stale, and failed states. Individual cells can be reviewed, corrected, and computed again. Inputs, engine, and run context remain traceable as the provenance of a generated value.

Tables belong to a project and persist across individual chats. Approved assistants can search, read, and update records through bounded table tools without direct SQL write access. For authoritative data in ERP, CRM, or local systems, an integration or Outpost remains the better path. Data Tables is in beta, and business-critical extractions need human review.