GitHub Copilot app Integration

GitHub Copilot app

The GitHub Copilot app is GitHub's desktop app for agentic coding. The Dash0 agent plugin adds an extension to it that sends an OpenTelemetry trace for every turn. Each trace records token usage, the model, tool and MCP calls, and sub-agent activity. Dash0 links those sessions to the pull requests they help ship. Engineering leaders and platform teams see in one place how widely the Copilot app is used, what it costs, and what it delivers.

Use Cases

Cost visibility: Track spend by user, team, and model so you can understand and control what the Copilot app costs.
Adoption tracking: Follow active users, new adopters, and usage distribution to see how the Copilot app rolls out across your engineering teams.
Productivity signals: Correlate sessions, tool calls, and assisted pull requests to understand the impact the Copilot app has on your team's output.

Overview

The Dash0 agent plugin includes an extension for the GitHub Copilot app. The extension joins each session, reads its events, and sends one trace per turn, with token usage per model, tool and MCP calls, and sub-agent activity. Dash0 prices the tokens to show cost. It leaves prompt and response text out unless you turn it on. See Data collected and privacy. The AI Coding section in Dash0 then shows what the Copilot app costs, how widely your teams use it, and what it helps ship.

The Copilot app and Copilot CLI read the same configuration file, so you configure Dash0 once for both. To set up the CLI, use the GitHub Copilot CLI integration.

Prerequisites

The GitHub Copilot app, installed and signed in. The extension runs on macOS, Linux, and Windows, on amd64 or arm64.
A Dash0 auth token with ingest permissions.
Your name and email in your git configuration. The extension uses them to attribute activity to users:

Install the extension

Ask Copilot in the app to install the extension, and give it the link to the repository:

Configure

Start a new session and send any prompt. The extension starts with the first prompt and posts dash0: telemetry is not active ... Run /dash0-configure. to the session timeline. Then run the configure skill:

When prompted, enter your Dash0 connection details:

OTLP URL: {{endpoint_otlp_http}}
Auth token: {{token_ingestion}}
Dataset (optional): {{dataset}}

For the privacy options (omit_io, omit_user_info), see Data collected and privacy. For help choosing a dataset, see the dataset documentation.

Attribute Usage by Team (Optional)

Dash0 breaks Copilot app activity down by user and model without extra setup. To break it down by team as well, enter a team name when you run /dash0-configure. The extension sends it as dash0.team.name on every span.

Verify

Start a new session in the Copilot app and send a prompt that uses a tool.
The session timeline shows dash0: connected. The extension starts with the first prompt, so the line appears after it.
Within a few seconds, the trace appears in the AI Coding section of Dash0. Each turn has a chat <model> span with token usage, an execute_tool <Name> span per tool call, and an invoke_agent <agent> span per sub-agent.

If no traces arrive, add debug: true and debug_file: /tmp/dash0-copilot-app-debug.log to ~/.copilot/dash0-agent-plugin.local.md. The extension then writes every span it sends to that file as a [dash0:trace] line. If the lines are there but Dash0 shows nothing, check the endpoint and token.

Uninstall

Delete ~/.copilot/extensions/copilot-app/ and start a new session. Keep ~/.copilot/dash0-agent-plugin.local.md if you still use the Copilot CLI plugin, which reads the same file.

Data collected and privacy

The extension sends spans over OTLP/HTTP to the endpoint you configure. It authenticates with your auth token and sets the Dash0-Dataset header when you configure a dataset. Before it runs the binary from GitHub Releases, it checks it against the release checksums.

The extension always sends:

The developer's git user.name and user.email, or the OS account name when git has no name set.
The working directory, the repository, the branch, and the commit SHA.
The model, token usage, tool names, MCP server names, skill names, and sub-agent names.
For commands run through the bash tool, only the binary name (for example git or npm), not the full command line.
Pull request, issue, and commit references found in tool output.

By default the extension does not send prompts, responses, tool inputs, or tool outputs. With omit_io at its default of true, it replaces them with <REDACTED>, and a failed turn reports only its error category. With omit_io: false, it sends that content truncated at 16 KB per attribute. It does not scrub secrets, so file contents, command output, and anything pasted into a prompt reach Dash0 unchanged.

With omit_user_info: true (default false), the extension hashes user.name, drops user.email, and replaces your home directory in the working directory with ~. With omit_identity_fallback: true, it drops the OS account name instead of reporting it. Set these options in the YAML frontmatter of ~/.copilot/dash0-agent-plugin.local.md (%USERPROFILE%\.copilot\dash0-agent-plugin.local.md on Windows).

To turn the extension off in one repository, commit .copilot/dash0-agent-plugin.local.md with enabled: false in the YAML frontmatter. The app runs each session in a fresh worktree, so an uncommitted file has no effect. The project file replaces the global file in that repository, and it also turns off the Copilot CLI plugin there.

Technology

GitHub Copilot app

Overview