Use AI
Calls AI for each row to extract, enrich or cleanup.
Last updated
Was this helpful?
Calls AI for each row to extract, enrich or cleanup.
Last updated
Was this helpful?
Find prompt examples
The Use AI node helps you run an AI model on your table, row by row. You write a prompt, pick a model, and the node returns the results for every row in new columns.
Describe what you want the AI to do for each row in the prompt box. Use @ColumnName to pull in values from your table.
Example:
Select the AI model you want to use from the drop-down menu. Not all models support web search
Turn on web search if your prompt needs fresh information from the internet. If the chosen model does not support web search, this option will be disabled.
Set the search depth to Low, Medium, or High:
high Most comprehensive context, highest cost, slower response.
medium (default) Balanced context, cost, and latency.
low Least context, lowest cost, fastest response, but potentially lower answer quality.
Add as many output columns as you need. For each column, set its name, type (String, Integer, Boolean, etc), and optional description to help AI.
You can use “Generate from prompt” to let the AI suggest columns based on your prompt.
If you don’t add any columns, the node will automatically create them based on the AI’s first response.
Switch to JSON schema view if you want to see or edit all columns as JSON.
You can run the node on all rows, just one row, or selected rows. An alternative way to get the result for a specific row is to find any output column in the table preview and click the "Generate row" link.
The node works row by row, passing each row’s values to the prompt and writing back the results.
After the run, you can manually edit, add, or delete values in the output columns. You can also re-run the node for any specific rows directly from the table, without affecting other rows.
Test your prompt
Start by running the node on one or a few rows to see if you get the desired output. Adjust your prompt or output columns before running it on the whole table. We recommend always using this option first. When you are ready, press the "All rows" button to return values for all remaining rows. It will not refill rows already containing the result.
One row at a time
The node works with only one row at a time. It cannot calculate aggregations, compare, or group rows.
Changing output columns
If you want to change the output columns after a run, reset the results.
Web search and credits
Using a web search will use more credits. Only turn it on if your prompt actually needs it.
Model limits
Each model has its limit for how much text it can process. Try a shorter prompt or pick a different model if you hit an error.
Show only affected columns.
Use the "Show only affected columns" toggle below the table to show only the columns used in the prompt or set as output columns and hide all other columns.
