CELF AI Settings¶
This section describes the screens related to various settings for CELF AI.
Generative AI Action Settings Screen¶
This is the screen for configuring settings for Generative AI actions.
The various settings specified on this screen are applied to all Generative AI actions used within that app.
Tip
If settings are configured for an individual action, those settings take precedence.
Open the 'Generative AI' tab and click the settings icon in the upper right corner of the tab.
In the 'Generative AI settings' dialog that opens, you can perform the following settings.
- Adjusting the request results … please refer to the table listing Generative AI Action Parameters.
- Knowledge management… please refer to How to use Knowledge.
- Supplementary information management… please refer to Managing Table Supplementary Information.
- Credit usage log … please refer to Output a Credit Usage Log.
Generative AI Action Parameters¶
There are the following four parameters for adjusting request results.
| Parameter name | Description |
|---|---|
| Randomness in selecting answer options | Specify the degree of randomness in selecting output options for answers. Higher settings will increase randomness. |
| Range to retain as final answer option | Specify the range for selecting the final candidate from answer output options. Higher settings will include a broader scope of content. |
| Suppressing word recurrence | Adjust the degree to which word repetition in text to be avoided. Higher settings reduce the repetition of the same expression. |
| Suppressing word reuse | Adjust the degree to which the reuse of words that have appeared once in text is avoided. Higher settings compose text with a more diverse selection of words. |
Output a Credit Usage Log¶
When you perform a test run using the 'Test run' button on the 'Action' ribbon menu, you can output information about the credits consumed by Generative AI actions to a log file.
The log records information such as the app name, sheet name, action set name, step number, action name, and number of credits consumed, allowing you to check afterward which action consumed how many credits.
The following credit-consuming actions are subject to logging.
In the 'Generative AI settings' dialog, select the 'Output a log during test runs' checkbox under 'Credit usage log' to enable log output.
Attention
- Logs are output only during test runs executed from the 'Test run' button on the 'Action' ribbon menu. Logs are not output when running a published app or when executing from 'Test run' in the 'Action set settings' dialog.
- For 'Launch Data Aggregation Agent', logs are output on a per-chat basis.
- This setting is saved per app.
- By default, log output is disabled (OFF).
A log file is created each day with the name
genai-YYYYMMDD.log, and that day’s logs are appended to it.Clicking 'Open the log output folder' opens the folder where the log files are stored.
Check Usage Status of Generative AI Actions¶
There is a monthly limit on the number of credits that can be used for generative AI actions.
You can check the usage status of Generative AI actions from the 'Usage status' view.
For the 'Usage status' view, please refer to Check usage status.
- [1] Monthly usage overview
You can check this month’s usage status of Generative AI actions.
Item Name Description Available credits This indicates the number of credits that can be used per month. The upper limit varies depending on your contract details. The number of credits consumed varies depending on the content of the executed Generative AI action. If there are multiple Generative AI actions in an action set, credits will be consumed accordingly. Available amount of knowledge data for sync This indicates the available capacity for Knowledge. The capacity varies depending on your contract details. - [2] Monthly usage status
- You can check the number of credits consumed and the amount of knowledge data used for each month.
Estimated Number of AI Runs¶
An estimate of roughly how many times you can perform each task if you use up a full month’s worth of credits (200,000).
"Consumption: Low" through "Consumption: Very high" are categories indicating how many credits are consumed; they are shown together with the model name in the AI model options.
| Task | Consumption: Low | Consumption: Medium | Consumption: High | Consumption: Very high |
|---|---|---|---|---|
Short requests
Brief answers to text of up to about 800 characters, such as checks or classification
|
Approx. 22,000-40,000 | Approx. 6,500-12,000 | Approx. 1,000-3,300 | Approx. 500 |
Summarizing / writing
Summarize or create text of about 1,000-2,000 characters
|
Approx. 9,500-15,000 | Approx. 2,500-4,800 | Approx. 400-1,300 | Approx. 200 |
Knowledge use
Generate an answer based on about 3,000 characters of internal knowledge
|
Approx. 7,400-15,000 | Approx. 2,500-4,400 | Approx. 360-1,200 | Approx. 180 |
Data aggregation agent
Instruct data aggregation in plain language, such as "aggregate sales by month"
|
— | — | Approx. 530 | — |
File reading (OCR)
Read PDFs or images such as scanned documents (1-5 A4 pages)
|
Approx. 3,500-7,000 | Approx. 1,100-2,100 | Approx. 170-560 | Approx. 84 |
Note
Actual consumption varies with the amount of input text, attachment size, answer length, and the amount of knowledge referenced.
Knowledge-based answers and the data aggregation agent may count as multiple AI calls per operation.
The data aggregation agent runs on a dedicated AI model, so its consumption is constant regardless of the selected model.