OpenTelemetry (OTel) is an open source observability framework. It provides a standard way to collect telemetry events and metrics and export them to compatible observability tools. For more information, see What is OpenTelemetry? on the OTel website.
When you enable OTel monitoring, you can send data from users' Copilot clients to an OTel-compatible backend. This lets you analyze agent sessions and understand agent usage across your enterprise.
What data does Copilot send?
Copilot sends three types of data:
- Traces show the flow of an agent session and connect each step, including model calls and tool use. For example, a trace can show an agent calling a model, using the
readFiletool, and calling the model again to produce a response. - Metrics are numeric measurements that help you identify patterns over time. For example, token usage metrics track the number of input and output tokens used in model calls.
- Events record individual actions at a specific point in time. For example, an edit feedback event records whether a user accepted or rejected an agent edit.
By default, the data does not include directly identifying attributes, prompts, responses, tool arguments, tool results, or raw policy details. Enterprise administrators can independently enable the following capture classes through managed settings:
- Identity adds the authenticated Copilot account name, local operating-system account, and machine name. This lets you attribute activity without collecting conversation or tool content.
- Prompts include user, system, and developer prompt content and associated model input.
- Responses include assistant response and supported reasoning content.
- Tool arguments include raw arguments, commands, tool schemas, and identifiers for tools, MCP servers, plugins, and hooks.
- Tool output includes tool results and execution error details.
- Policy detail includes unbounded policy information such as raw sandbox filesystem paths. Bounded policy posture and outcomes remain available when this class is disabled.
Fine-grained capture controls are supported in Copilot CLI, the GitHub Copilot app, and VS Code sessions using Agent Host. VS Code's local agent harness supports identity capture and the legacy all-or-nothing content capture setting, but does not independently apply the five content capture classes.
Captured content may contain sensitive information, including source code, file contents, user prompts, commands, tool results, and workspace or user-profile paths. Enable only the classes required for your use case.
Enabling OpenTelemetry
To collect OTel data from users' Copilot clients:
- Set up an observability backend. Choose a secure backend that supports the OpenTelemetry Protocol (OTLP). Some backends can receive OTLP data directly. For other backends, deploy an OpenTelemetry Collector to receive, process, and forward the data. For examples, see Use with observability backends in the VS Code documentation.
- Configure users' clients. Set configuration values to enable OTel in each client and configure it to send data to your OTLP endpoint by configuring the endpoint, headers, and authentication token.
- Interpret data. Use the collected data to analyze sessions and identify trends. For an example implementation, see Monitor AI coding agents with Grafana in the Microsoft documentation.
Configuring users' clients
Enterprises can enforce OTel configuration across supported clients with managed settings. Explicit managed values take precedence over local settings, SDK configuration, and environment variables. The telemetry property includes keys for enabling and routing OpenTelemetry and independently controlling identity and sensitive content capture.
For more information, see Getting started with enterprise-managed settings and Enterprise managed settings.
Client documentation
For more details of how OpenTelemetry is used across clients, see:
- Monitor agent usage with OpenTelemetry in the VS Code documentation
- OpenTelemetry instrumentation for Copilot SDK
- GitHub Copilot CLI command reference