Introduction
Over the past couple of years, many of us have been utilizing Generative AI interfaces and co-pilots to enhance our communication, conduct research, and summarize complex information. AI-based agents are digital entities created to autonomously derive insights from data and execute actions. Agents are focused on accomplishing a specific outcome without the needfor constant human intervention. Within the realm of IT Operations and observability, these agents can sift through massive datasets and perform tasks such as supplementary data collection, issue validation, and problem correction.
Fabrix.ai, formerly known as CloudFabrix, is now adding Agentic AI capabilities into Robotic Data Automation Fabric. Key components of agentic AI framework are:
- Agent Orchestration and Lifecycle Management
- AI Guardrails
- Managing Data and Action Privileges for Agents
- Visibility and Observability of Agents
- Agent Quality Control and Assurance
- Reasoning LLMs
Let’s take an example use case:
Anomaly Detection and see how Agentic AI differs from ML based approach.
- ML-Based Anomaly Detection:
- Utilizes machine learning algorithms to identify outliers in data based on statistical patterns or pre-established rules.Requires substantial amounts of training data.Primarily focuses on identifying anomalies, leaving interpretation and response to human operators or integration with other systems (e.g., AIOps).Limited ability to explain the cause of anomalies, assess their impact, or suggest corrective measures.
- Often requires continuous retraining to adapt to changing data patterns.
- Agentic AI-Based Anomaly Detection:
- Employs AI agents to detect anomalies, often with less reliance on large datasets.Agents can reason about the causes of anomalies and their potential impact on the system.Can autonomously recommend or execute corrective actions.Adapts and learns from past anomalies, dynamically adjusting its response strategies.
- Offers greater potential for explainability and autonomous decision-making in response to anomalies.
Agent Orchestration & Lifecycle Management
Agentic AI within the Fabrix.ai platform empowers users to create agents by either defining new task descriptions or utilizing predefined templates. The Agentic AI framework employs Large Language Models (LLMs) to translate these descriptions into a task graph, which outlines the sequence of tasks, dependencies, and conditions for the agent.
The Agentic AI orchestrator executes the task graph, while the AI Guardrails module ensures that the AI adheres to the defined task and doesn’t deviate. The task graph typically encompasses the following node types:
- Query Node: Executes queries against the Data Fabric to retrieve data or aggregations.
- Generation Node: Employs LLMs to generate queries, summaries, insights, and recommended actions.
- Decision Node: Utilizes LLMs to make decisions based on data and previous generation results, determining the next steps within the graph.
- Action Node: Leverages the Automation Fabric to perform automated actions, such as opening tickets, sending notifications, triggering other agents, or executing data retrieval and remediation tasks.
The Agent Lifecycle management encompasses the following actions:
- Review: Enables users to review and refine the task graph to align with objectives.
- Dry Run: Allows users to simulate agent actions without actual execution to preview outcomes.
- Test: Facilitates agent execution for a limited duration or iterations on specified test data to verify outcomes.
- Deploy: Enables agent deployment onto the platform, either on a schedule or triggered by events.
- Decommission: Takes agents offline.
AI Guardrails
AI-based agents, designed to perform specific tasks autonomously, necessitate guardrails to prevent unintended or harmful actions. This is especially crucial for agentic AI platforms that enable users to create new agents by describing the desired task.
AI Guardrails can be categorized into three major areas:
- Standard or common sense guardrails:
These are fundamental guardrails designed to prevent the AI from generating outputs that contain profanity, use language unsuitable for a professional setting, or promote harmful or illegal activities. - Industry / Domain specific guardrails:
These guardrails are tailored to specific industries or domains. They encapsulate industry best practices and ensure that the AI’s output complies with relevant regulations. For example, an AI agent in the healthcare sector would need to adhere to patient privacy laws. - Enterprise / Environment specific guardrails:
These guardrails are specific to an individual enterprise or environment. They enforce company policies and values, ensuring that the AI’s output aligns with the organization’s culture and standards.
While some AI models may have been pre-trained with certain guardrails, it is risky to assume that they are comprehensive and prevent all harmful activity. A robust agentic AI framework must ensure that agents never engage in unwanted behavior.
In our AI Fabric, we base our guardrails on well-established benchmarks and allow our customers to refine them further to meet their specific requirements. Anytime an agent is created, AI Fabric ensures that it complies with these AI guardrails. Furthermore, AI Fabric monitors every interaction between the agent and LLM to ensure ongoing compliance with the established guardrails.
Managing Data Access for AI Agents
Within our Data Fabric, data is organized into Streams, Datasets, and Graph DB collections. AI agents, possessing user-like personas, can have access restrictions applied to limit their scope to specific streams or datasets. Granular control can be further implemented to restrict access to particular slices of data within a stream.
Managing Action Privileges for Agents
Autonomous agents will have the capability to execute numerous actions. Therefore, it is crucial to manage their privileges and limit the set of actions they can perform. The Agentic AI platform will ensure that these action privileges cannot be overridden by the AI itself, maintaining control and security.
Visibility and Observability of Agents
Fabrix.ai Storyboards provide insights into the operation and performance of agents by visualizing agent workflows, tracking progress, and identifying bottlenecks or areas for optimization. This enhanced visibility enables better management and coordination of agent activities, ultimately leading to improved efficiency and effectiveness. Additionally, Storyboards capture detailed reasoning used by the LLM, which helps users understand the AI’s decision-making process and refine it if necessary.
Agent Quality Control and Assurance
Agent Quality Control is crucial for maintaining the reliability and consistency of AI agents. This process involves continuous monitoring of agent behavior, identifying and rectifying errors or performance issues, and implementing improvements to optimize agent effectiveness and prevent operational disruptions.
The Agent Quality Control module leverages Large Language Models (LLMs) to oversee agent performance and ensure that they are successfully achieving their intended outcomes.
Additionally, the system incorporates user feedback on both specific agent executions and overall agent performance. This feedback loop enables the Agent Quality Control module to identify emerging issues in new executions and further enhance agent performance.
How can Fabrix.ai assist me if I already have a significant amount of data in a data platform?
The RDAF has pipelines that enable data collection and ingestion from various data sources. This data can then be stored within RDAF or forwarded to other data platforms.
However, if your data already resides in platforms such as Splunk, Opensearch, Elasticsearch, or Arango DB (Graph Database), Fabrix.ai can leverage this data directly for visualization and Agentic AI capabilities without requiring re-ingestion into RDAF. For additional databases that we natively integrate with, please refer to our technical documentation. We are continuously adding more.
To achieve this seamless integration, simply add the API credentials for your chosen platform (whether on-premises or cloud-based) into RDAF. Once these credentials are added, you can immediately start utilizing the data from any of those platforms.
Copyright (c) 2025 Fabrix.ai Inc. All rights reserved.
