Uploads

What uploads are for

Uploads let you bring a CSV or XLSX file into an Agent conversation for analysis, comparison, or modeling context.

Use uploads for:

• One-time spreadsheet analysis.
• Comparing a file to warehouse data.
• Providing seed data for a demo or draft model.
• Sharing a mapping table or business rule sheet.
• Giving the Agent context that is not yet connected as a managed source.

[picture 1. Agent input with uploaded XLSX file staged for the next message]

Supported file types

The Agent upload control accepts:

• .csv
• .xlsx

If you have another file type, convert it to CSV or XLSX before uploading.

Upload workflow

1. Open Agent.
2. Click the paperclip icon.
3. Select a CSV or XLSX file.
4. Wait for the upload to appear above the input.
5. Describe how the Agent should use the file.
6. Send the message.

Uploaded files are associated with the current conversation. If you refresh the page, Stackless can recover staged uploads for that conversation.

Tell the Agent what the file is

Always describe the source and intent of the upload. This avoids confusion with existing warehouse sources.

Good examples:

This uploaded workbook is a one-time customer cohort list. Compare it to existing customer data and create a table artifact of missing matches.

This CSV is seed data for a demo. Use it to create a draft analysis, but do not assume it is a connected source.

This spreadsheet should match the existing customer source. Use customer_id as the join key and show unmatched rows.

Ambiguous uploads

If Stackless is unsure whether a sheet is seed data or close to an existing source, the Agent UI may ask you to clarify. In your message, state whether the upload should be treated as:

• New seed data.
• A one-time analysis file.
• A file to match against an existing source.
• A replacement or correction for existing source data.

What the Agent can create from uploads

The Agent can use uploads to create:

• Table artifacts.
• Briefs or findings.
• Comparison summaries.
• Data-quality checks.
• Draft transformation ideas.
• Dashboard inputs, if the uploaded data is appropriate.

For recurring dashboards, a one-time upload is usually not enough. The data should be connected as a source or modeled into a governed Stackless path.

Privacy and sensitivity

Treat uploads as data shared with the Stackless workspace. Avoid uploading direct personal identifiers or confidential files unless your organization has approved that use.

If a table artifact includes sensitive columns, ask the Agent to mask them:

Create a table artifact from this upload, but mask email, phone, and address columns.

Troubleshooting uploads

If an upload does not work:

• Check the file type is CSV or XLSX.
• Check that the file is not empty.
• Try removing password protection or unsupported workbook features.
• Use shorter sheet names and clear column headers.
• Split very large files before upload if the Agent cannot inspect them reliably.

If the Agent misclassifies the file, correct it explicitly in the next message.