Skip to content

Enabling AI Services

To enable AI services in your deployment, provide the AI provider configuration details in the ai_provider.yaml file. This configuration must include the code name of the AI provider and any provider-specific options. Refer to the Links and Resources section below for detailed configuration options for each supported provider.

Currently, the deployment supports the following AI providers and their respective code names:

  • AWS Bedrock (aws)
  • Azure OpenAI (azure)
  • Ollama (ollama)

Tip

By default, the deployment looks for the ai_provider.yaml file in the root of the inventory directory. You can override this location by setting the ai_provider_file variable in the inventory file. A sample ai_provider.yaml file is provided in the root of the installation package.

Sample ai_provider.yaml
  ai_provider:
    name: aws
    model: anthropic.claude-haiku-4-5-20251001-v1:0
    temperature: 0.7
    max_tokens: 100000

Authentication and Permissions

For information on configuring cloud authentication and IAM/RBAC permissions required for AI services, see the AI Services Permissions sections in the Cloud Integrations guide.

Log Analysis with AI

VOR Stream supports AI-powered log analysis for each run. This feature uses your configured AI provider to summarize and interpret logs, helping you quickly identify issues and understand what happened during the run.

After a run completes, you can view the log analysis by clicking the "Explain Logs" button in the log viewer for that run.

Configuring Log Analysis Limits

  • By default, the system analyzes up to 500 log entries per run.
  • You can change this limit by setting the run_log_analysis_max_logs variable in your Ansible inventory file or Django config.
  • Increasing the limit provides deeper analysis but may impact performance and cost (depending on your AI provider).

Example (hosts.ini file)

[django:vars]
run_log_analysis_max_logs=1000

Customizing AI Prompts

AI prompts can be customized during deployment using Ansible templates. This allows you to tailor the AI's behavior for your specific environment without modifying the application code.

How It Works

  1. Default prompts are defined in {vor_stream_package}/django/template/ai_prompts/ as Jinja2 templates (.j2 files).
  2. During deployment, if an AI provider is configured, Ansible deploys these templates to the Django application.
  3. After deployment, prompts are located in {vor_root}/django/src/ai/prompts/.

Prerequisites

AI prompts are only deployed when an AI provider is configured in ai_provider.yaml. If no AI provider is set up, the prompt deployment task is skipped.

Prompt Template Format

Each prompt consists of two separate text files:

  • {prompt_name}_system.txt: System instructions for the AI assistant
  • {prompt_name}_user.txt: User message template with variable placeholders

Variables in the user template are specified using curly braces: {variable_name}

Example:

my_prompt_system.txt:

You are an expert AI assistant.
Provide clear and concise answers.

my_prompt_user.txt:

Please analyze the following data for user {user_id}:

{data}

Customizing Prompts for Your Deployment

To customize AI prompts for your deployment:

  1. Navigate to your installation package directory:

    cd <installation_package_directory>/django/template/ai_prompts/

  2. Edit the desired template file(s):

    # For example, customize the log summarization system prompt
vi summarize_logs_system.txt.j2

  3. Make your changes. For example:

    You are an expert analyst for financial risk applications.
Focus specifically on:
- Data quality issues
- Calculation errors
- Regulatory compliance warnings

Provide recommendations that align with risk management best practices.
Keep responses concise and actionable for business users.

  4. Deploy the changes:

    cd ../../../  # Back to ansible directory
ansible-playbook django/playbook.yaml -i hosts.ini

  5. The Django service will automatically restart with the new prompts

Advanced Customization with Jinja2

Since these are Jinja2 templates, you can use any Ansible variables or Jinja2 features for advanced customization:

Using Ansible variables: 

You are an AI assistant for {{ inventory_hostname }}.
Data center: {{ datacenter | default('dc1') }}

Conditional logic: 

{% if custom_ai_tone is defined and custom_ai_tone == 'formal' %}
Provide formal, business-oriented recommendations.
{% else %}
Provide clear, practical recommendations.
{% endif %}

Custom variables: Define custom variables in your inventory file (e.g., hosts.ini) and reference them:

[django:vars]
ai_focus_area=financial risk management

Then use in your prompt:

You are an expert in {{ ai_focus_area | default('data analysis') }}.