What Are Agents?
Agents in ADM are specialized AI components designed to perform specific data management tasks. Each agent is equipped with a focused set of capabilities, access permissions, and tools that enable it to act autonomously or collaboratively within ADM’s multi-agent ecosystem.
An agent typically has:
- Defined responsibilities within a particular data domain
- Access to targeted tools and datasets relevant to its function
- Expertise in performing domain-specific analysis or actions
- Collaboration capabilities to work seamlessly with other agents
Agent Architecture
Agent Types and Capabilities
ADM includes several classes of agents categorized by their purpose: Discovery, Monitoring, Analysis, and Notification. When you submit a complex query—such as _“Create a data quality policy for the customer table and notify me when it fails”_—multiple agents may work together through collaborative execution:
- The Catalog Agent retrieves schema and metadata.
- The Data Quality Agent recommends validation rules.
- The Policy Creation Workflow builds the policy configuration.
- The Incidents Agent sets up alerting and notifications.
ADM’s orchestration layer automatically routes the request, invokes the required agents (often in parallel), aggregates their results, and synthesizes the final response.
Discovery Agents
| Agent | Purpose | Capabilities | Use Case | Example Query |
|---|---|---|---|---|
| Catalog Agent | Discovers and catalogs data assets | Discovery | Find datasets, retrieve metadata, view schema details | “Show me all tables containing customer data.” |
Monitoring Agents
| Agent | Purpose | Use Case | Example Query |
|---|---|---|---|
| Data Quality Agent | Monitors data quality standards and metrics | Track quality scores, check policy executions | “Which data quality policies failed in the last 24 hours?” |
| Schema Drift Agent | Detects schema changes and drift patterns | Identify schema changes and breaking impacts | “Has the customer table schema changed recently?” |
| Data Drift Agent | Monitors data distribution changes over time | Detect distribution shifts or anomalies | “Show me tables with significant data drift.” |
| Data Cadence Agent | Tracks data freshness and arrival frequency | Monitor freshness and SLA compliance | “Which tables haven’t been updated today?” |
| Pipelines Agent | Monitors data pipeline executions | Track pipeline health, history, and performance | “Show pipeline failures from yesterday.” |
Analysis Agents
| Agent | Purpose | Use Case | Example Query |
|---|---|---|---|
| Reconciliation Agent | Compares data across systems for consistency | Identify discrepancies and validate counts | “Compare customer counts between source and target.” |
| Web Search Agent | Searches external sources for supporting information | Retrieve documentation or best practices | “What are the industry standards for data quality policies?” |
Notification Agents
| Agent | Purpose | Use Case | Example Query |
|---|---|---|---|
| Incidents Agent | Manages and tracks data incidents | Create, track, and resolve incidents | “Create an incident for the failed customer_orders policy.” |
How Agents Work Together
Collaborative Execution
When you submit a query that requires multiple steps, ADM coordinates agent collaboration automatically.
Example:
“Create a data quality policy for the customer table and notify me when it fails.”
| Agent | Function |
|---|---|
| Catalog Agent | Retrieves the customer table schema and metadata. |
| Data Quality Agent | Analyzes the structure and recommends quality rules. |
| Policy Creation Workflow | Builds and applies the policy configuration. |
| Incidents Agent | Configures notifications for policy failures. |
Orchestration Process
- ADM’s Routing Layer determines which agents are relevant.
- Agents execute in parallel whenever possible to reduce latency.
- The Orchestrator aggregates and reconciles their outputs.
- The final, unified response is synthesized and displayed to the user.
Enabling and Configuring Agents
To view or configure agents:
Navigate to Settings > Agents.
The Agents page lists all available agents with:
- Agent name and icon
- Description of purpose
- Capability badges (e.g., Discovery, Monitoring, Analysis, Notification)
Use the search bar to filter by agent name or capability.
Enabling Agents
- Locate the desired agent and toggle the Enable switch to ON.
- The agent becomes active immediately, no restart required.
- Most agents use system defaults, but advanced users can configure:
- Monitoring frequencies
- Alert thresholds
- Data source or integration parameters
Note Advanced configuration should be performed by an administrator to maintain system consistency.
Agent Orchestration in Action
During query execution, ADM provides real-time visibility into which agents are active through the Agent Thinking Panel. This view displays:
- The agents currently in use (e.g., Catalog, Data Quality, Reconciliation)
- Step-by-step progress indicators
- Execution timings for each agent
- Actions taken and their outcomes
Benefits of Orchestration Transparency
- Understand how ADM interprets and executes your request
- Verify that the correct agents were invoked
- Troubleshoot unexpected outputs
- Learn better prompting strategies by observing how queries are processed
The orchestrator ensures that all relevant agents contribute efficiently, merging their outputs into a coherent and actionable response.