Skip to main content
Tables — referred to internally as datagrids — are structured data stores that AI agents can query and populate during conversations. Unlike a Knowledge Base, which stores free text for semantic search, a table stores data in columns and rows with defined types, like a lightweight database directly accessible by the agent.

Core Concepts

ConceptDescription
Table (Datagrid)A data structure with named and typed columns, organized into rows. Each table belongs to a workspace and can be connected to one or more agents.
ColumnDefines a field in the table. Each column has a name (up to 50 characters), description (3 to 250 characters), data type (String, Number, Boolean), and optional settings such as semantic search and required field.
RowA record in the table. Each row contains values for each defined column. Rows can be inserted manually through the interface, imported via CSV, or created by the agent during conversations.
Datagrid ToolA capability made available to the agent for interacting with the table. The available actions are: semantic_search, similarity_search, insert_row, and update_row. Each action is activated individually per table.
Column Semantic SearchWhen enabled on a String column, values are vectorized using the text-embedding-3-small model (1,536 dimensions) and indexed for similarity search — the agent finds rows by meaning, not just exact match.

Why Tables Matter

  1. Real-time data without external integration: the agent accesses tables directly during a conversation. No webhooks, APIs, or integrations need to be configured for the agent to query a product catalog or register a lead.
  2. Separation of structured and unstructured knowledge: Knowledge Bases are ideal for free text — manuals, policies, FAQs. Tables are ideal for data with a defined structure — catalogs, schedules, inventories, records. Using each resource for what it was designed for improves agent accuracy.
  3. Writing during the conversation: unlike Knowledge Bases (read-only for the agent), tables allow the agent to insert and update data in real time — the agent collects information from the user and records it in a structured way in the table, without human intervention.
  4. Updates independent of the prompt: data changes in the table without needing to edit the agent prompt. A price updated in the table is immediately available for the agent on the next query.
Typical use cases where tables outperform other approaches:
  • Product catalog with price, availability, and description
  • Professional or event schedule with time slots and vacancies
  • Lead records collected during conversations
  • Inventory with real-time quantity tracking