Instana AI assistant (Public Preview)

The Instana AI assistant is a conversational interface that uses natural language to help you analyze and understand your observability data. Currently available as a public preview in SaaS environments and Self-Hosted environments, this feature uses IBM WatsonX.ai to transform complex monitoring information into clear, actionable insights through simple conversational queries.

Instana AI assistant capabilities

The AI assistant acts as an intelligent bridge between you and your Instana environment, enabling you to:

  • Query observability data naturally: Ask questions in plain English rather than learning complex query syntax
  • Extract insights quickly: Receive immediate answers about your infrastructure, applications, and incidents
  • Analyze multiple data sources: Access information across application calls, infrastructure components, and events or issues through a unified interface
  • Visualize results dynamically: View outputs in tables or download them for further analysis

Core advantages of Instana AI assistant

  • Reduced learning curve: Work in natural language without needing to learn Instana's query syntax or API details
  • Accelerate troubleshooting: Receive diagnostic answers within seconds
  • Contextual intelligence: Enable the AI to interpret your intent and apply the right filters and aggregations automatically
  • Flexible data exploration: Switch between different technology stacks

Supported technologies

You can query the following technologies through the AI assistant:

Application performance
Application calls analyze service-to-service communications, latency, and throughput.
Infrastructure components
  • DB2 database: Monitor database performance metrics and overall health
  • Kubernetes pod: Monitor pod-level metrics, resource usage, and status
  • Kubernetes deployment: Monitor deployment health, replica counts, and rollout status
  • IBM MQ queue: Analyze queue depth, processing rates, and messaging performance
  • JVM runtime platform: Monitor Java Virtual Machine metrics, garbage collection activity, and memory usage
Events and incidents
Incidents or issues help investigate alerts, incidents, and their root causes.

Prerequisites

You can use the Instana AI assistant only if the following prerequisite conditions are met:

SaaS environments

For SaaS environments, Instana provides default LLM gateways. No extra configuration is required to use this capability. You can configure separate gateways for your own runtime if necessary. For more information, see Configuring LLM gateways.

This feature uses the feature flag feature.instana.chat.enabled and feature.coordinator.ai.agent.enabled, which is enabled (set to true) by default.

Self-hosted environments

For self-hosted environments, you must configure LLM gateways to use this capability. For more information, see Configuring LLM gateways.

In addition, you must enable the feature.incident.ai.summarization.enabled, feature.ai.gateway.enabled, and feature.coordinator.ai.agent.enabled feature flags (set to true). For more information, see Self-hosted optional flag AI assistant.

Getting started

To use the AI assistant, complete the following steps:

  1. From the navigation menu in the Instana UI, select Events.
  2. Click the AI icon to open the chat interface. Chat entry point showing the AI icon in the Instana Events section

The chat window appears. You can type your queries in natural language.

Using the prompt library

The AI assistant includes a library of predefined queries to help you learn effective query patterns.

Opening the prompt library:

  1. In the AI assistant chat interface, click the Open prompt library button. Open prompt library button in the AI assistant chat interface
  2. Click one of the following tabs:
    • Application: Queries related to application performance and service calls
    • Infrastructure: Queries for infrastructure components and resources
    • Events: Queries about incidents, alerts, and issues

Using predefined prompts:

Each prompt in the library is a template with editable placeholders:

  1. Select a pre-filled prompt from the library. Pre-filled prompt selected from the prompt library
  2. Replace the variables or placeholders with your specific values as shown in the following example:
    • service-name (for example, "payment-service")
    • time-range (for example, "last 30 minutes", "past 24 hours")
    • metric-threshold (for example, "response time > 500ms")
    • resource-identifier (for example, pod names, deployment names)
  3. Select the Click to send message arrow to execute your query.

Example of prompt usage:

Original prompt template:

Show me all calls to [service_name] with errors in the [time_range]

Customized query:

Show me all calls to payment-service with errors in the last 2 hours

Understanding query processing

You can write effective queries by understanding how the AI assistant processes your queries.

Step 1: Understanding natural language

When you submit a query, the AI assistant analyzes your input to:

  • Identify the technology: Determines the category of a query, such as application calls, infrastructure components, or events
  • Extract intent: Understands what you want to know (for example, list resources, show metrics, or find errors)
  • Recognize parameters: Identifies time ranges, filters, thresholds, and other constraints

Example:

  • Query: "Show me Kubernetes pods with high CPU usage in the last hour"
  • Technology identified: Kubernetes pod
  • Intent: List resources with filtering
  • Parameters: CPU metric, threshold (high), and time range (1 hour)
Step 2: Generating API payload

The AI assistant translates your natural language query into a structured API call:

  • Constructs API request: Builds the appropriate Instana API payload
  • Applies filters: Adds tag filters, time ranges, and metric thresholds
  • Formats response: Specifies how data is aggregated and displayed (table, chart, and so on)
Step 3: Displaying data retrieval

The processed query is executed against your Instana environment:

  • Fetches data: Retrieves information from the Instana API
  • Formats results: Presents data in table format
  • Provides context: Shows how the query is interpreted and if any assumptions are made

Multi-agent architecture

The AI assistant uses a multi-agent architecture powered by a coordinator agent that intelligently routes queries to specialized agents based on your request.

How the coordinator works:

The coordinator agent acts as the central orchestrator of the AI assistant:

  • Query analysis: When you submit a query, the coordinator analyzes your natural language input to understand the intent and context.
  • Agent selection: Based on the query type, the coordinator routes your request to the appropriate specialized agent (such as the API agent for data retrieval).
  • Response coordination: The coordinator manages the interaction between agents and ensures you receive a coherent, formatted response.

Currently, the AI assistant only uses the API agent for querying Instana data, with the coordinator managing the overall interaction flow. This architecture enables future expansion to support additional specialized agents as new capabilities are introduced.

Working with query results

After your query is processed, the AI assistant provides rich, interactive results.

Data table view
The table view interface provides sortable columns, allowing you to click column headers to organize the data as needed. It also includes scrollable content, making it easy to navigate through large result sets. In addition, you can access detailed information, enabling you to view comprehensive metrics and attributes for deeper analysis.
Export to CSV
The download option enables you to export query results for offline analysis, providing a full data export that includes all columns and rows returned by the query. The exported file is delivered in a compatible format, allowing it to open in Excel, Google Sheets, or other analytical tools for further review.
Search within results
The filtering feature allows you to quickly locate specific entries within large result sets by narrowing the displayed data. It provides real-time filtering, updating results instantly as you type, and supports case-insensitive search, ensuring matches are returned regardless of capitalization.
Data visualization
Results are presented in a tabular format that allows you to analyze and understand your observability data effectively. The table view provides sortable columns and scrollable content for easy navigation through large result sets, with access to detailed metrics and attributes for comprehensive analysis.

Understanding query interpretation:

The AI assistant provides transparency about how it processed your query. The interpretation panel provides an interpretation summary, outlining how the AI understood your query. It also identifies any assumptions made, such as default values that were automatically applied (for example, a time range defaulting to 30 minutes). In addition, it details the applied filters, listing all filters and constraints included in the underlying API call.

Example:

Query: "Show me slow database calls"
Interpretation:
- Technology: Application Calls
- Filter: Database calls only
- Threshold: Response time > 1000ms (default)
- Time range: Last 30 minutes (default)

Large Language Model (LLM) details

The AI assistant uses the following LLM for query processing:

  • Model: openai/gpt-oss-120b
  • Provider: IBM watsonx.ai
  • Capabilities: Natural language understanding, intent recognition, and API payload generation

The AI processing flow begins by receiving your query, interpreting the natural-language input to understand the request. It then analyzes the context, recognizing observability concepts and Instana-specific terminology to accurately determine intent. Using this understanding, the system generates structured output by creating the necessary API calls with appropriate filters and parameters. Finally, it provides explanations, offering transparency into how the query was interpreted and translated into actionable operations.

Best practices

Writing effective queries:

  1. Be specific about time ranges
    • Good: "Show me errors in the last 2 hours"
    • Less effective: "Show me recent errors"
  2. Include relevant filters
    • Good: "List Kubernetes pods in the production namespace with high memory usage"
    • Less effective: "Show me pods"
  3. Specify thresholds when relevant
    • Good: "Find application calls with response time greater than 500ms"
    • Less effective: "Show me slow calls"
  4. Use technology names clearly
    • Good: "Show me IBM MQ queue depth for order-queue"
    • Less effective: "Show me queue information"

Query refinement:

You can begin by starting with a broad query to obtain an initial understanding of the available data. Based on the results, you can then refine your query with additional filters to narrow the scope and focus on specific details. For more complex requests, you might also use the prompt library as a starting point, leveraging predefined examples to guide the creation of more advanced queries.

Leveraging default values:

The AI assistant applies sensible defaults when information is missing:

  • Time range: 30 minutes (if not specified)
  • Thresholds: Industry-standard values for "high", "low", "slow", and so on
  • Sorting: Most relevant or recent items first

You can override these defaults by specifying your preferences in the query.

Limitations and considerations

Public preview status:

As a public preview feature, the AI assistant might occasionally show inaccurate results when handling complex queries, and it remains subject to ongoing changes and improvements based on user feedback. In addition, it does not yet support all Instana features and data types, with broader coverage planned over time.

Scope of query:

This feature is currently available only in SaaS environments and is not supported for on-premises Instana deployments. Its functionality is limited to the specific technologies outlined in this document, and all queries are subject to Instana's existing data-retention policies, which determine how much historical data you can access.

Performance:

Complex queries might require additional processing time, and when a query returns a large volume of data, the system shows it in pages instead of all at once. In addition, real-time data might experience slight delays before appearing in the results.

Troubleshooting

AI assistant doesn't understand my query
  • Try rephrasing using terminology from the prompt library.
  • Be more specific about the technology and time range.
Results don't match expectations
  • Review the query interpretation section to see how the AI understood your request.
  • Add more specific filters or constraints.
No results returned
  • Verify the time range includes the period you're interested in.
  • Check that the specified resources exist in your environment.
  • Broaden your query filters.
Chat interface not visible
  • Ensure you're in the Events section of the Instana UI.
  • Verify that the Incident AI summarization feature is enabled.
  • Check that you're using a SaaS deployment.
Raw tool call format appears in response

Symptom: Responses displaying raw tool calls in the format [TOOL_CALLS]instana_nl_to_api{"query": "show me closed incidents from the last 5 days", "technology": "events"}

Cause: This is an issue that has occurred when using the Mistral LLM model. This issue did not occur consistently, and has been resolved.

Solution: If you encounter this issue consistently, consider switching to an alternative LLM model through your LLM gateway configuration.

Feedback and support

As this is a public preview feature, user feedback is especially valuable. You are encouraged to report any issues or unexpected behavior through your standard IBM support channels, share suggestions for enhancements or new capabilities, and provide examples of queries that did not perform as expected.

Related documentation

Summary

The Instana AI assistant transforms the way you work with observability data by simplifying data access through natural-language queries, accelerating troubleshooting with immediate, actionable insights, reducing operational complexity by automatically managing API calls and data formatting, and enhancing productivity with pre-built prompts and interactive results. Begin exploring your Instana environment today and experience the power of conversational AI.