Error analysis using AI

Available from version 26.3: This topic describes how to use AI error analysis to review synchronization errors, and get suggested resolution steps.

Overview

AI error analysis helps you diagnose synchronization errors faster. It compares the error details to similar issues, and suggests practical next steps.

Use AI error analysis when a connection run fails, and you want a starting point before collecting trace files or contacting support.

Set up AI error analysis

This section describes how to set up AI error analysis.

To set up AI error analysis:

  1. Install the pgvector extension for your PostgreSQL version.

    • Windows: Download the appropriate pgvector package for your PostgreSQL version, and extract the vector files into C:\Program Files\PostgreSQL\<PostgreSQL_Version>\.

    • Linux: Install the package that matches your operating system and PostgreSQL version. For example: yum install -y pgvector_16 --disablerepo=epel

  2. Run the following SQL command in PostgreSQL:

    CREATE EXTENSION vector;

  3. Load the provided knowledge base JSON into the exceptionembeddings table, and run the relevant script with the JSON file path:

    The knowledge base JSON files are provided separately.

    • Windows: mfcErrorHandlingUsingAI.bat <json file path>

    • Linux: mfcErrorHandlingUsingAI.sh <json file path>

    Pass the JSON file path as an argument, or update the script to include the file path in the command.

  4. Use the mfcFullRestClient utility to enable AI error analysis:

    java -jar mfcFullRestClient.jar -h localhost:8081 -c <user-name>,<password> -setGlobalPropertyValue -propertyName llm.integration.ai.audit.analysis.enabled -propertyValue true

    On Linux RHEL 9, restart the Connect service after you enable the audit analysis property. This step is not required on Windows or Linux RHEL 8.

  5. Use the mfcFullRestClient utility to set the AI audit analysis service URL:

    java -jar mfcFullRestClient.jar -h localhost:8081 -c <user-name>,<password> -setGlobalPropertyValue -propertyName llm.integration.ai.audit.analysis.url -propertyValue https://generativelanguage.googleapis.com/v1beta/openai/chat/completions

    If you use a Gemma-based deployment, set the appropriate service URL.

  6. Use the mfcFullRestClient utility to set the AI model:

    java -jar mfcFullRestClient.jar -h localhost:8081 -c <user-name>,<password> -setGlobalPropertyValue -propertyName llm.integration.ai.audit.analysis.modelname -propertyValue gemini-2.5-flash

    If required, replace gemini-2.5-flash with the appropriate Gemma model.

  7. Use the mfcFullRestClient utility to set the API authentication token:

    java -jar mfcFullRestClient.jar -h localhost:8081 -c <user-name>,<password> -setGlobalPropertyValue -propertyName llm.integration.ai.audit.analysis.token -propertyValue <API_TOKEN>

    Replace <API_TOKEN> with a valid API authentication token.

Analyze an error

Before you analyze an error, make sure the AI error analysis setup is complete.

To analyze a synchronization error:

  1. Open Audit logs for the current connection.

  2. Expand the failed items, and look for the AI icon in the failure stack trace row.

  3. Click the AI icon to view the resolution steps.

  4. Review the suggested resolution steps.

  5. Run one iteration, and check Connection Messages and Audit to verify the result.

When analyzing an error, keep the following in mind:

  • Suggestions are intended to help with common or known error patterns.

  • Results may vary depending on the error details and available context.

  • If the issue continues, gather trace-level logs and contact support.

See also