Analyze data

note

Deprecated: Analyze data is a legacy tool. It remains available for existing workflows but is no longer actively developed. For new data-analysis workloads, use Autonomous Agents.

The Analyze data tool runs Python in an isolated sandbox to calculate, transform, and visualize structured data for an agent. Use it with uploaded files, linked files, or structured output from earlier steps. It doesn't change the source files.

With Analyze data, you can:

• Explore dataset columns, values, and structure
• Calculate counts, sums, averages, trends, and other statistics
• Filter, join, group, or reshape tabular data
• Create summaries, charts, and downloadable files

Inputs

Configure Analyze data in Agent builder with instructions and at least one data input. Be specific about the calculation and expected output.

Input or setting	What to provide	Example
Analysis instructions	Describe the calculation, transformation, or visualization	Group open tickets by priority and calculate the average age for each group
Input data	Reference an uploaded agent input, a linked file, or structured output from an earlier step	Use the CSV from [[Search Jira with JQL]]
Output format	State how the tool should present or save the result	Return a short summary and save all filtered rows as a CSV file

Analyze data supports structured formats such as CSV, XLS, XLSX, XLSM, Google Sheets, and JSON. See Data analysis overview for format details and file guidance.

Outputs

Analyze data returns a Markdown summary and can generate tables, charts, or downloadable files. The result becomes step output in the agent's memory. Add a Respond step to present the findings and links to generated files to the user. Later steps can access the result through their memory settings or a direct step reference.

When a result contains more than 1,000 rows, Analyze data saves the full result to a file and returns a summary in the step response. Ask for a CSV or spreadsheet output when the user needs every row. Don't ask a later Respond step to reconstruct the complete dataset from the summary.

Usage examples

• Ask, "What columns and date ranges does this spreadsheet contain?"
• Group opportunities by stage, then calculate record counts and total value for each stage
• Filter retrieved tickets to a date range, select the required fields, and save the rows as a CSV file
• Create a chart that compares monthly support volume by product area

Prerequisites

• Enable data analysis for your organization in Admin console → Platform → Assistant → Settings → Data Analysis. See Enable data analysis
• To use uploaded files, enable file upload and give the running user access to the files. See File upload setup
• Add a retrieval step before Analyze data when the agent must get records from a connector

Configuration

Add Analyze data to an agent:

1. Open the agent in Agent builder and add a step.
2. Select Analyze data under By Glean.
3. Enter explicit analysis instructions.
4. Reference an agent input or an earlier step output that contains the data.
5. State the expected output, such as a summary, chart, CSV file, or spreadsheet.
6. Add a Respond step or another downstream step that uses the analysis.

For reliable results, name the relevant columns, filters, calculations, and output format. When you need data from an earlier non-sequential step, reference that step directly instead of loading unrelated step output into memory.

Limitations

• Analyze data processes the data you provide. It doesn't search connectors or retrieve records by itself
• Upload limits depend on the model. See Upload limits
• Glean might truncate indexed spreadsheet content before it reaches the tool. For large or wide files, upload the source file or reference it directly instead of passing text extracted by a Read document step
• Complex sheets with merged cells, irregular tables, embedded charts, or diagrams can reduce accuracy
• The sandbox has execution time, memory, CPU, and library limits. It doesn't allow network access
• Analyze data returns results with more than 1,000 rows as a generated file with a summary in the step response

Troubleshooting

Confirm that data analysis is available for your organization. If the agent uses uploaded files, confirm that your admin has also enabled file upload.

Add an uploaded file, link to a file, or reference structured output from an earlier step. Confirm that the step's memory settings include that output, or reference the step directly in the instructions.

Check whether the tool received the source file or truncated text from an earlier step. For large spreadsheets, upload the file or reference the file directly in Analyze data. Include the required sheet names, columns, filters, and calculations in the instructions.

If the result has more than 1,000 rows, use the generated file for the complete dataset. Configure the downstream Respond step to present the summary and file link rather than reproduce every row in its response.

Reduce the input size, remove unused columns, split the work into smaller batches, or simplify the requested transformation. Then run the step again.

Frequently asked questions

Yes. Reference structured output from a search, query, or read step. Use the step's memory settings or a direct step reference to make that output available.

Yes, within the file limits for your organization's model. Identify each file and explain how the tool should join or compare them.

No. It reads the provided data in a sandbox. The sandbox returns transformed data as new output or a generated file.

No. The sandbox doesn't have network access. Retrieve the required data in an earlier agent step, then pass it to Analyze data.

See also

• How agents work — understand step output, memory, and direct references
• Limits and exhaustive retrieval best practices — retrieve and process large record sets
• File security — review sandbox isolation and resource controls
• Build agents with atomic tools — combine retrieval, analysis, and response steps