Product Update: Jun 12 | Tables: Structured Data for AI Agents

Announcements
,
1

Your AI Agents can now treat spreadsheets as real, structured data. Say hello to Tables.

:bar_chart: Tables: a new data source for AI Agents

Until now, you could only upload structured data in the CSV file format to Documents. That only allowed your AI Agents to match rows using semantic search (similarity in meaning). That works fine for articles and FAQs, but not for product catalogs: exact codes weren’t matched, multi-column questions returned half-matches, and requests that required data comparison and calculations were not possible.

Tables fix that:

  • Exact lookups: ask for an ID, part number or SKU, and your AI Agent will fetch the right data, every time.
  • Multi-column filters: “red running shoes in size 42” returns only rows that match every condition, not just one of them.
  • Ranges and sorting: “shelves under 30cm deep,” “cheapest first.”
  • Calculations: totals, counts, averages, minimums, and maximums, worked out on the spot.
  • Partial text match: search by part of a name, like “wireless” to surface every wireless model you sell.
  • Unique values: “what sizes do you stock?” returns the full list with no duplicates.

:card_index_dividers: Setting up a table

Find Tables under Automations > Data Sources, or per Agent under Knowledge > Tables.

  • Two ways to start: upload a CSV or XLSX (Excel format now supported – one worksheet per upload), or build one from scratch in the spreadsheet-style editor.
  • Type confirmation: the system detects each column’s type (number, text, boolean, date, email, URL) and lets you confirm before importing.
  • Descriptions drive accuracy: the AI reads your table description to decide when to reference it, and your column descriptions to pick the right fields. Thus, adding clear and specific table and column descriptions is recommended for the best results.

Tip: descriptions can not only tell your AI Agent how to read a column, but also enrich the data, e.g. “prices are in GBP, excluding VAT” or “these are relative paths (slugs), pre-pend https://yoursite.com/ to return a full URL.”

:arrows_counterclockwise: Keeping tables up to date

When your data changes, re-import from the table’s edit view:

  • Replace: overwrite everything with a fresh file (e.g. today’s full catalog export).
  • Append: add new rows to what’s already there (e.g. this week’s orders).

We will be adding the ability to automatically update data in Tables in future updates.

:bulb: When to use Tables and when to use Documents?

  • Use Tables for structured data: catalogs, price lists, inventory, specs, orders.
  • Use Documents for prose: FAQs, policies, help articles, product write-ups, etc.

Your AI Agent can reference and combine information from both for a single reply, so you don’t need to load the same data twice.

:pushpin: Things to Note:

  • Each table requires a description. It tells your AI Agent when to query the table.
  • Column types are set at creation. To change a column’s type, delete it and add a new one with the desired type.
  • One worksheet per upload. Specifically relevant for multi-sheet XLSX files. You can only add one sheet at a time.
  • Limits: up to 50,000 rows and 100 columns per table, 1,000 characters per cell, and 50 MB per upload.