Skip to main content

AI Tools Guide

This guide explains how Cloudaware customers can leverage AI tools to interact with this repository and the data within the Cloudaware CMDB.

Integration Requirements​

To effectively use an AI agent with this repository, the following prerequisites must be met:

  • Tool Usage: The AI agent must be capable of using tools, such as reading files and executing shell commands. An example of such a tool is the Gemini CLI.
  • Large Context Window: The model should support a large context window (over 128K tokens). Google's Gemini family of models, with up to 1M input tokens, meets this criterion. We recommend using Gemini 1.5 Pro.
  • Instruction Following: The model must be proficient at following instructions, especially within a large context. gemini-2.5-pro is significantly better at this than gemini-2.5-flash.

How It Works​

Cloudaware provides an agents.md file that bootstraps the AI agent with initial instructions.

Based on the user's request, the AI agent can access further documentation within this repository, including:

By exploring the repository, the AI agent can utilize:

  • Information about all available types to select the correct one based on user queries.
  • Detailed type descriptions, including available fields (e.g., AWS EC2 Instance).
  • Available Compliance Engine policies, located in the /ce directory.

The system also relies on the AI's built-in knowledge of:

  • Salesforce Object Query Language (SOQL)
  • BigQuery Query Language
  • AWS, Azure, GCP, and other cloud platforms

Example Use Cases​

The following are tested use cases that demonstrate the AI agent's ability to understand and correctly query data in the Cloudaware CMDB. These examples are illustrative; AI agents can use similar methods to perform more complex tasks.

Querying Data in the CMDB​

Example Query:

How many running EC2 instances do I have?

AI Agent's Process:

  1. Determines the correct object type and its API name: CA10__CaAwsInstance__c.
  2. Inspects available fields in either type.json or the generated documentation.
  3. Concludes that CA10__stateName__c must be running.
  4. Filters out deleted objects by ensuring CA10__disappearanceTime__c is null.
  5. Constructs a SOQL query.
  6. Converts the SOQL query to a BigQuery query using repo-manager soql convert.
  7. Executes the BigQuery query using the gcloud command and provides the answer.

Similar questions that can be answered:

  • How many security groups open port 22 to the public internet?
  • What is the largest EBS volume in my environment?
  • Show me a list of all IAM users without MFA enabled.

Handling Customer-Specific Fields​

Example Query:

How many EC2 instances in the us-east-1 region are owned by johndoe@mycompany.com?

AI Agent's Process:

  • The agent follows the same general process as above but correctly infers that "owned by" likely refers to the owner tag, which is stored in the caTag_owner__c field.

Writing Compliance Policies​

Example Request:

Write a policy that marks S3 buckets as non-compliant if "MFA delete" is not enabled.

AI Agent's Process:

  1. Looks up all relevant API names for types and fields.
  2. Consults the repository structure documentation to understand the necessary components for a new policy.
  3. Creates the required folder descriptors.
  4. Creates the policy descriptor.
  5. Generates the logic file using the appropriate operations to implement the policy.