Note
Access to this page requires authorization. You can try signing in or changing directories.
Access to this page requires authorization. You can try changing directories.
Use this FAQ to understand transparency, reliability expectations, limitations, and how to interpret outputs from Azure Copilot Observability Agent.
For data retention, residency, access controls, and governance settings, see Data, privacy, and governance FAQ for Azure Copilot Observability Agent.
What is Azure Copilot Observability Agent?
Azure Copilot Observability Agent helps you investigate and analyze telemetry data such as logs, metrics, traces, and alerts across Azure resources by using a conversational experience.
It uses AI and machine learning to automate investigation tasks, translate natural language questions into queries, interpret and visualize results, and suggest mitigation steps based on available telemetry. For an overview, see Azure Copilot Observability Agent overview.
What can Azure Copilot Observability Agent do?
The Observability Agent provides a chat experience where you can ask questions about observability data, trigger deep investigations, and adjust investigation time and scope.
During analysis and investigation, the agent can:
- Interpret your intent and relevant scope.
- Map related resources and dependencies.
- Generate relevant queries to fetch observability data.
- Detect anomalies.
- Correlate findings and potential causes.
- Summarize and visualize results.
- Suggest mitigation options.
- Explain reasoning progress during the investigation.
- Suggest follow-up prompts.
The investigation process is surfaced step by step so you can review findings as they appear. At the end, the agent provides a summary and the supporting findings.
What is the intended use of Azure Copilot Observability Agent?
Operations and development teams use the agent to analyze Azure-monitored resources and applications. It helps them identify problems faster, determine likely root causes, and evaluate possible remediation steps.
How was Azure Copilot Observability Agent evaluated?
Microsoft evaluated the experience through multiple channels, including:
- Product telemetry trends.
- Customer feedback from in-product channels.
- Survey results and text feedback.
- Customer calls.
- Automated synthetic tests in test environments and shared workspaces.
What are the limitations and how can I reduce their impact?
- Lack of monitoring data: The investigation process relies on monitoring data, such as metrics and logs. If you don't set up Azure resources and Application Insights components to provide even standard logging, the investigation process might not be successful. This limitation applies to both existing (quick mode) investigations and new agentic (deep thinking mode) investigations.
- Lack of large language model (LLM) token capacity: If many investigations run at the same time, request volume can exceed available token capacity and delay responses.
What operational factors support effective and responsible use?
- Set up Azure resources and applications with standard monitoring instrumentation to output sufficient monitoring data for investigations.
- Adjust investigation time range and resource scope as needed.
How do I provide feedback?
Provide feedback through:
- Thumbs up and thumbs down, and the associated text box.
- Azure feedback experience.
- A survey available in the Azure portal banner.
Are the results reliable?
Azure Copilot Observability Agent is designed to generate the best possible suggestions based on the data and context it can access. Microsoft repeatedly assesses and calibrates it to provide reliable insights and evidence. However, like any AI-powered system, output might not always be perfect. Carefully evaluate and validate results generated by Azure Copilot Observability Agent before taking action in your Azure environment.
Are the outputs authoritative?
No. The agent provides assistive insights based on available telemetry data. You're responsible for validating outputs before taking action.
What data does the agent access?
The agent accesses Azure observability data and related resource data within the scope of the signed-in user's permissions. It operates within existing access management controls, such as Azure role-based access control (Azure RBAC), Microsoft Entra Privileged Identity Management, Azure Policy, and resource locks.
Can the agent access data I'm not authorized to view?
No. The agent operates under Azure RBAC and can access only data that you are already authorized to access.
How does the agent use data from my Azure environment?
Azure Copilot Observability Agent analyzes Azure observability data to reason over it and find operational issues related to scoped resources.
What data does the agent collect?
Azure Copilot Observability Agent doesn't use user data, prompts, or responses to train or improve underlying AI models. To improve Microsoft products and services, it might collect usage engagement data, such as the number of sessions, session duration, selected skills, and feedback, subject to the Microsoft Privacy Statement and applicable consent requirements. The agent doesn't collect personally identifiable information (PII).
What should I do if I see unexpected or offensive content?
The development of Azure Copilot Observability Agent is guided by AI principles and the Responsible AI Standard and prioritizes preventing irrelevant or offensive output. However, as with any AI feature, unexpected results might still appear. Microsoft continuously works to improve this technology to prevent such content.
How current is the information the agent provides?
Azure Copilot Observability Agent uses the latest observability data available it finds. Data freshness depends on ingestion and processing, so recently generated telemetry might appear with a delay.
What are the fairness considerations?
Fairness is a core part of Azure Copilot Observability Agent development. Consistent performance across different scenarios and input data types is critical. The development team evaluates the system for fairness by checking reliability signals and potential incorrect outputs. They take measures to prevent harmful generated text and ensure fallback options are in place when AI encounters issues such as timeouts or service unavailability.
Where can I read more about Responsible AI standards?
For more information, see the Microsoft Responsible AI Standard.