Skip to main content
Create Collection dialog with CSV import option
The fastest way to bring structured data into the foundation is to upload a CSV file. Collections parses your file, detects column names and types automatically, and shows you a preview before anything is created, the structured-data equivalent of uploading documents into a knowledge base.

​
Uploading a CSV

1

Start a new collection

Click New Collection, then choose Import a CSV.
2

Add your file

Drag your .csv file into the drop zone, or click it to open the file picker. Files up to 50 MB are accepted.
3

Review and adjust the schema

Once parsed, the columns open in an editor pre-filled with the detected names and types. Rename columns, change a type, mark which are required, and flag columns to index or make unique. A live preview shows the resulting table.
4

Create

Click Create. The collection is created with your schema, then the rows are imported in the background through the batch endpoint, with a progress bar (sent / total).

​
How Type Detection Works

Collections scans a sample of your rows and assigns each column the most specific type that fits all its values: Whatever is detected, you can override any column’s type in the schema editor before creating. A few rules to know:
  • Mixed columns fall back to text. If a column contains 12, 15, and N/A, it becomes text; clean the stray values first if you need numeric aggregations.
  • Empty cells are allowed. Columns with missing values are marked nullable; empty cells are shown as a dash.
  • Numbers are strict. Only plain decimal notation is parsed (no scientific notation, no 0x prefixes). Leading-zero values like 007 are kept as text so identifiers are preserved.
For dates, export in ISO format (YYYY-MM-DD) from your source tool. Localized formats like 25/02/2026 are not recognized as dates and will import as text.

​
File Requirements

Excel files (.xlsx) are not supported directly; export your sheet as CSV first.
The parser handles messy real-world files: semicolon/tab delimiters (French/Excel exports), a UTF-8 BOM, CRLF or LF line endings, quoted fields with commas or line breaks, blank lines, and rows with missing trailing values (filled with empty). It does not auto-detect files without a header row (the first row is treated as the header).

​
Import Errors

​
Adding Data Later

Importing at creation is not the only way to fill a collection:
  • Add rows and columns in the table. In the Data tab, use Add row to insert a record and Add column to extend the schema; see Working with Data
  • Agents can write records during conversations, using the collection’s insert and update tools; see Using Collections with Agents
  • The API accepts batch inserts of up to 5,000 records per call via the bulk endpoint (file imports and large loads use it automatically, chunked); see API Reference
Schemas are flexible: records added later can introduce new fields, and the collection’s columns adapt without a migration.
Available soon: the capability store will add connectors (Salesforce, HubSpot, and more) to sync data into collections directly from your business tools, and hooks to pre-process records as they arrive.

​
Next Steps

Working with Data

Browse and manage your imported records

Using Collections with Agents

Let agents read and write your data