GitHub Copilot Integration

GitHub Copilot already exports OpenTelemetry data: agent turns, LLM calls, tool executions, token usage, and the AI credits each call used. Traccia ingests that data natively on the hosted platform. Point Copilot at Traccia, work as usual, and open the dashboard to see which tools ran, which models answered, what it would cost at API prices, and how turns roll up into conversations. It works with Copilot Chat in VS Code and the GitHub Copilot CLI.

1Get a Traccia API key

Sign in to app.traccia.ai, open Settings, then API Keys, and create an ingestion key. Copy the key. You will paste it into Copilot's settings in the next step.

2Configure Copilot in VS Code

Open your user settings (Command Palette, then Preferences: Open User Settings (JSON)) and add the following. Replace YOUR_TRACCIA_API_KEY with your key. The endpoint is the standard OTLP base path: Copilot appends /v1/traces, /v1/metrics, and /v1/logs itself.

HTTP (recommended)

{
  "github.copilot.chat.otel.enabled": true,
  "github.copilot.chat.otel.exporterType": "otlp-http",
  "github.copilot.chat.otel.otlpEndpoint": "https://api.traccia.ai/otel",
  "github.copilot.chat.otel.headers": { "Authorization": "Bearer YOUR_TRACCIA_API_KEY" },
  "github.copilot.chat.otel.captureContent": true
}

Agent mode: set environment variables

The settings above trace Copilot Chat's own model calls. Agent mode runs in Copilot's bundled runtime, which reads only environment variables and only when VS Code starts. Set these, quit every VS Code window, and start VS Code from the same shell. Encode the space after Bearer as %20, as the OpenTelemetry spec requires.

export COPILOT_OTEL_ENABLED=true
export OTEL_EXPORTER_OTLP_ENDPOINT=https://api.traccia.ai/otel
export OTEL_EXPORTER_OTLP_HEADERS="Authorization=Bearer%20YOUR_TRACCIA_API_KEY"
export OTEL_INSTRUMENTATION_GENAI_CAPTURE_MESSAGE_CONTENT=true
code .

$env:COPILOT_OTEL_ENABLED = "true"
$env:OTEL_EXPORTER_OTLP_ENDPOINT = "https://api.traccia.ai/otel"
$env:OTEL_EXPORTER_OTLP_HEADERS = "Authorization=Bearer%20YOUR_TRACCIA_API_KEY"
$env:OTEL_INSTRUMENTATION_GENAI_CAPTURE_MESSAGE_CONTENT = "true"
code .

To set them once on Windows, save them as user environment variables, then quit and reopen VS Code. Any launch, including the Start menu, picks them up. They also apply to any other app that reads the standard OTEL_EXPORTER_OTLP_* variables.

setx COPILOT_OTEL_ENABLED "true"
setx OTEL_EXPORTER_OTLP_ENDPOINT "https://api.traccia.ai/otel"
setx OTEL_EXPORTER_OTLP_HEADERS "Authorization=Bearer%20YOUR_TRACCIA_API_KEY"
setx OTEL_INSTRUMENTATION_GENAI_CAPTURE_MESSAGE_CONTENT "true"

On macOS and Linux, VS Code opened from the Dock or an app menu does not read your shell profile. Add the export lines to ~/.zshrc or ~/.bashrc and start VS Code with code from a terminal.

If your Copilot Chat has no github.copilot.chat.otel.headers setting, remove that line from settings.json. The environment variables above supply the key.

gRPC

Copilot Chat's own calls can also go over gRPC on port 4317. Agent mode always uses OTLP/HTTP through the environment variables. For the general gRPC reference, see Platform OTLP Ingestion.

{
  "github.copilot.chat.otel.enabled": true,
  "github.copilot.chat.otel.exporterType": "otlp-grpc",
  "github.copilot.chat.otel.otlpEndpoint": "https://api.traccia.ai:4317",
  "github.copilot.chat.otel.headers": { "Authorization": "Bearer YOUR_TRACCIA_API_KEY" },
  "github.copilot.chat.otel.captureContent": true
}

3Configure the Copilot CLI (optional)

The GitHub Copilot CLI is configured with environment variables. It exports over OTLP/HTTP (JSON by default, or protobuf with OTEL_EXPORTER_OTLP_PROTOCOL=http/protobuf); it has no gRPC option. Sign in once with copilot login, then set these in the shell you run copilot from, or in your shell profile:

export COPILOT_OTEL_ENABLED=true
export OTEL_EXPORTER_OTLP_ENDPOINT=https://api.traccia.ai/otel
export OTEL_EXPORTER_OTLP_HEADERS="Authorization=Bearer%20YOUR_TRACCIA_API_KEY"
export OTEL_INSTRUMENTATION_GENAI_CAPTURE_MESSAGE_CONTENT=true
copilot

$env:COPILOT_OTEL_ENABLED = "true"
$env:OTEL_EXPORTER_OTLP_ENDPOINT = "https://api.traccia.ai/otel"
$env:OTEL_EXPORTER_OTLP_HEADERS = "Authorization=Bearer%20YOUR_TRACCIA_API_KEY"
$env:OTEL_INSTRUMENTATION_GENAI_CAPTURE_MESSAGE_CONTENT = "true"
copilot

4Run Copilot and open Traccia

Use Copilot as you normally would. In VS Code, open Copilot Chat in agent mode; in a terminal, start the CLI:

copilot

Ask it to do real work: read files, run commands, edit code. Copilot exports spans in batches every few seconds, and each turn is complete once the turn finishes.

In Traccia, go to Traces and look for the agent GitHub Copilot (id github-copilot). Click any trace to open the full timeline. Your first trace usually appears within seconds of a turn completing.

What gets captured

• An invoke_agent span per agent turn. In the CLI each turn is its own trace. In VS Code each chat session is one trace, rooted at VS Code's vscode.agent_host.session span, with one invoke_agent per prompt inside it.
• A chat span per LLM call, with the model and input, output, cached, and reasoning tokens.
• An execute_tool span per tool call, with the tool name, and its arguments and result when content capture is on.
• Prompts and responses on each LLM span when content capture is on.
• Subagents appear as nested spans under the same Copilot agent, not as separate agents.
• Surface. Each trace is tagged IDE (VS Code) or CLI.
• Git context (VS Code). When Copilot Chat reports the repository and branch, they are shown on the trace. The CLI does not send git details.

VS Code may also emit hook spans for Copilot hooks, and the CLI emits a plan span when it plans a task. Both appear in the timeline.

Sessions

Copilot tags every turn with a stable gen_ai.conversation.id. Each prompt you send in the same conversation creates a new turn that shares that ID, and Traccia groups those turns into one session. In the CLI, resuming a conversation with copilot --resume continues the same session.

In Traccia, turn on Group by session on the Traces page to collapse a conversation into one row:

• Session row: total tokens, wall time, and estimated cost rolled up across all turns
• Turn rows: individual prompts expanded underneath
• Open: session detail page with every turn in order, plus GitHub Copilot and IDE or CLI badges
• Badges: turn count and tool retry count at a glance

A turn is marked as failed only when the agent turn itself errors. If a tool call fails inside a turn that otherwise succeeds, the error stays on that tool span and the turn shows Success with a tool retry badge.

Understanding cost

GitHub bills Copilot through your plan, not per token. Traccia therefore labels Copilot costs as estimates:

Agent identity

By default every turn rolls up under one agent, GitHub Copilot, whether it came from VS Code or the CLI. Subagents never create separate agents.

To split teams or environments, set your own agent.id and env as resource attributes:

"github.copilot.chat.otel.resourceAttributes": {
  "agent.id": "copilot-backend",
  "agent.name": "copilot-backend",
  "env": "production"
}

export OTEL_RESOURCE_ATTRIBUTES="agent.id=copilot-backend,agent.name=copilot-backend,env=production"

Roll out to your team

Give each team a Traccia workspace and an ingestion key scoped to that workspace so traces stay isolated, then distribute the settings above. If your organization manages Copilot's OpenTelemetry settings centrally, those managed settings replace each developer's personal ones, so configure the Traccia endpoint and key there.

Troubleshooting

Current limitations

• Copilot's metrics and log events are accepted and stored, but they do not have dedicated dashboards yet.
• Cost is an API-equivalent estimate, not your GitHub bill.
• The Copilot CLI does not send git repository or branch details.