October 16 – Release Notes

Discover the latest updates and improvements to Kognitos.

Release Date	October 16, 2025
Brain Version	main-d282ea7b52-43040
User Interface Version	main-2a7a040842-61936

New Features

API Endpoint Details

You can now click on Copy API Endpoint to see the v2 REST API endpoint for a given process, along with parameters such as Agent ID and Process ID. This update makes it easier to quickly access and copy process details.

Access this information by navigating to any process, clicking on the three-dot menu (⋮) in the top-right, and then selecting Copy API Endpoint.

Table Extraction from Excel with LLMs

You can now specify which LLM to use when extracting tables from Excel worksheets.

Use the openai model is for GPT models or the gemini model is for Gemini models. For example:

extract a table from the worksheet where
    the description is "Sales data table with monthly revenue and profit columns"
    the openai model is "gpt-5"

Convert JSON to Table

You can now convert JSON directly into a table, making it easier to work with structured data:

convert the jsons to a table

Enhancements

Browser Automation

Browser automation now supports file uploads and downloads. Files downloaded during browser automation tasks are automatically stored as the browser files, making them readily available for further processing.

We've also added a new procedure, to close a browser. While browsers automatically close after 4 hours, this procedure allows you to manually close specific browser instances as needed.

Table Extraction

Table extraction is now more accurate and reliable with several key improvements.

A new strict mode in the extract a table procedure matches prompts to formal table structures for improved header and value accuracy.

The system also uses a two-pass approach, first identifying headers and titles before extracting row data. When your prompts specify headers, they now guide extraction when aligned with the document content.

Resolved Issues

• Asking for User Input: Resolved an issue where automations would incorrectly pause to ask for user input instead of automatically retrying after certain transient errors, such as network failures.

• Providing Values: Resolved an issue where providing a value for one missing field in a document would incorrectly apply the same value to all other missing fields in the same request.

• Table Extraction: Resolved an issue causing table extraction failures for certain documents due to malformed data and complex structures.