Docs/Integrations/Coding Agent Telemetry

Coding-Agent Telemetry

AI coding agents export OpenTelemetry (OTLP) data. ANTS Platform ingests it and builds one trace per prompt, grouped into sessions and attributed to a user. Each trace carries the model, token counts, cost and tool calls.

ANTS Platform has dedicated parsers for three coding agents that export OTLP natively:

Claude Code
GitHub Copilot CLI
OpenAI Codex CLI

ℹ

Content capture is opt-in and off by default. Without the content settings, each tool sends usage metadata only: counts, models, durations and tool names. Prompt and response text is sent only after you turn on the content setting for that tool. One exception: Codex CLI always sends tool arguments and tool output (see the Codex CLI notes).

Two ways to set it up

Method	Best for	What you do
Automatic	Devices enrolled in Discover AI	Run the Discover AI enrollment script. It detects the coding agents on the device and configures them.
Manual	Individual developers, or devices without the enrollment script	Set the environment variables or config file shown under Manual setup.

How it works

Each tool exports OTLP over HTTP to one ingest endpoint, authenticated with a project key pair. ANTS Platform detects which tool sent the data and parses it into traces.

	Value
Endpoint (base)	`https://<your-ants-host>/api/public/otel`
Signal paths	`/v1/traces` and `/v1/logs` (most exporters append these to the base)
Content types	`application/x-protobuf` or `application/json`; gzip is accepted
Auth	HTTP Basic with a project public key and secret key

Rendering diagram…

Authentication

Use a project public key (pk-ap-...) and secret key (sk-ap-...). Project API keys are under Settings > API Keys. The Discover AI project keys under Settings > Deployment & Keys also work.

Join the two keys with a colon and Base64-encode the result:

bash

printf 'pk-ap-...:sk-ap-...' | base64

Send the value as an Authorization: Basic header:

bash

export OTEL_EXPORTER_OTLP_HEADERS="Authorization=Basic <base64(pk:sk)>"

⚠

The secret key grants ingest access to your project. Distribute it through managed configuration, not shared documents or repositories.

Automatic setup with the enrollment script

The Discover AI enrollment script (macOS and Windows) configures coding-agent telemetry as part of enrollment. See Deploying Agents & Extensions.

It detects each tool by its configuration folder in the user's home directory (.claude, .copilot or the copilot command, .codex).
Claude Code and Copilot CLI: it writes the OTLP endpoint, headers and user.id into a marked block in ~/.zshenv on macOS, or into the user's environment variables on Windows.
Codex CLI: it writes a marked [otel] block into ~/.codex/config.toml with trace_exporter = "none". If the file already has an [otel] section that ANTS did not write, the script leaves it unchanged.
It sets user.id to the user's email when the script can resolve one.
On projects that use device certificates, the tools send to a local relay run by the desktop agent, which forwards the data with the device certificate. No secret key is written into the tool configuration.
Re-running the script replaces its own block. The uninstall script removes it.

Turn on content capture

Content capture is set per project, per tool, under Settings > Deployment & Keys > Coding-agent telemetry > Capture content. All three tools are off by default. The setting is applied the next time the enrollment script runs on a device.

Tool	What the enrollment script adds when content capture is on
Claude Code	`OTEL_LOG_USER_PROMPTS=1`, `OTEL_LOG_TOOL_DETAILS=1`, `OTEL_LOG_TOOL_CONTENT=1`
GitHub Copilot CLI	`OTEL_INSTRUMENTATION_GENAI_CAPTURE_MESSAGE_CONTENT=true`
OpenAI Codex CLI	`log_user_prompt = true` (otherwise `false`)

User attribution

ANTS Platform attributes each trace to a user from the resource attribute user.id. It also accepts enduser.id and user.account_id.

⚠

Set user.id, not user.email. A resource-level user.email is ignored as an override, so a stale value in a shell profile cannot replace a tool's own login identity.

Tool	Identity the tool sends	How to get a readable user
Claude Code	`user.email` from the Claude login, on every span	Automatic. Set `user.id` only to override.
OpenAI Codex CLI	`user.email` and `user.account_id` from the ChatGPT login, on every event	Automatic. Set `user.id` only to override.
GitHub Copilot CLI	Only `enduser.pseudo.id`, a non-reversible hash	Set `user.id=<email>`. Without it, traces are attributed to the hash.

Manual setup

Set the values for each coding agent you use. Add environment variables to your shell profile (~/.zshrc or ~/.bashrc). For fleet rollout without the enrollment script, see Enterprise MDM rollout.

⚠

Set OTEL_SERVICE_NAME per tool. If OTEL_SERVICE_NAME="claude-code" is exported in a shell, any other agent started from that shell inherits it and can be attributed to the wrong tool. Run each agent in its own shell, or set OTEL_SERVICE_NAME for each tool as shown below.

Claude Code reads standard OTEL_* environment variables.

bash

export CLAUDE_CODE_ENABLE_TELEMETRY=1 export CLAUDE_CODE_ENHANCED_TELEMETRY_BETA=1 export OTEL_TRACES_EXPORTER=otlp export OTEL_METRICS_EXPORTER=otlp export OTEL_LOGS_EXPORTER=otlp export OTEL_EXPORTER_OTLP_PROTOCOL=http/protobuf export OTEL_EXPORTER_OTLP_ENDPOINT="https://<your-ants-host>/api/public/otel" export OTEL_EXPORTER_OTLP_HEADERS="Authorization=Basic <base64(pk:sk)>" export OTEL_SERVICE_NAME="claude-code" export OTEL_RESOURCE_ATTRIBUTES="deployment.environment=production,user.id=<your-email>"

Optional content capture. Leave these unset to send metadata only:

bash

export OTEL_LOG_USER_PROMPTS=1 # user prompt text export OTEL_LOG_TOOL_DETAILS=1 # tool names and input export OTEL_LOG_TOOL_CONTENT=1 # tool input and output export OTEL_LOG_RAW_API_BODIES=1 # assistant response text

ℹ

Assistant response text arrives only through OTEL_LOG_RAW_API_BODIES. Extended thinking is always redacted by Claude Code. Restart the terminal or Claude Code session after changing the variables.

Other tools

Claude Cowork: ANTS Platform also parses Claude Cowork OTel log events sent to /api/public/otel/v1/logs with the same Basic auth header.
Gemini CLI and Cursor have no dedicated parser. Gemini CLI cannot attach an auth header to its OTLP export, and Cursor has no native OpenTelemetry export.

Planned capabilities are listed on the roadmap.

What each tool can export

Some fields are never sent by a tool. In those cases the trace shows a [<Field> not exported by <Tool>] placeholder. This is expected and is not an ingestion error.

Field	Claude Code	Copilot CLI	Codex CLI
User prompt	✓ with `OTEL_LOG_USER_PROMPTS`	✓ with the content variable	✓ with `log_user_prompt`
Assistant text	✓ with `OTEL_LOG_RAW_API_BODIES` (thinking redacted)	✓ with the content variable	✗ never sent
Tool name	✓	✓	✓
Tool arguments	✓ with `OTEL_LOG_TOOL_DETAILS`	✓ with the content variable	✓ always sent
Tool result	✓ with `OTEL_LOG_TOOL_CONTENT`	✓ with the content variable	✓ always sent
Model, tokens, cost	✓	✓	✓

Environment variable reference

Tool	Key	Value
Claude Code	`CLAUDE_CODE_ENABLE_TELEMETRY`	`1`
Claude Code	`CLAUDE_CODE_ENHANCED_TELEMETRY_BETA`	`1` (enables the span view)
Claude Code	`OTEL_TRACES_EXPORTER` / `OTEL_LOGS_EXPORTER` / `OTEL_METRICS_EXPORTER`	`otlp`
Claude Code	`OTEL_LOG_USER_PROMPTS` / `OTEL_LOG_TOOL_DETAILS` / `OTEL_LOG_TOOL_CONTENT`	`1` to capture content; unset for metadata only
Claude Code	`OTEL_LOG_RAW_API_BODIES`	`1` to capture assistant response text
Copilot CLI	`COPILOT_OTEL_ENABLED`	`true`
Copilot CLI	`COPILOT_OTEL_ENDPOINT`	`https://<your-ants-host>/api/public/otel`
Copilot CLI	`OTEL_INSTRUMENTATION_GENAI_CAPTURE_MESSAGE_CONTENT`	`true` to capture content; unset for metadata only
Codex CLI	`[otel]` table in `~/.codex/config.toml`	`log_user_prompt`, `trace_exporter = "none"`, logs `exporter`
Claude Code, Copilot CLI	`OTEL_EXPORTER_OTLP_ENDPOINT`	`https://<your-ants-host>/api/public/otel`
Claude Code, Copilot CLI	`OTEL_EXPORTER_OTLP_HEADERS`	`Authorization=Basic <base64(pk:sk)>`
Claude Code, Copilot CLI	`OTEL_SERVICE_NAME`	`claude-code` or `github-copilot`
All	`OTEL_RESOURCE_ATTRIBUTES`	`deployment.environment=production,user.id=<email>`

Verify it's working

Confirm the configuration

bash

env | grep -E "CLAUDE_CODE_ENABLE_TELEMETRY|OTEL_|COPILOT_OTEL" cat ~/.codex/config.toml

Generate activity

Run a prompt in the tool, for example a question in Claude Code or a Codex command that calls a tool.

Check ANTS Platform

Each prompt appears as its own trace within a minute or two. Traces are grouped on the Sessions tab and attributed on Users.

Troubleshooting

Nothing shows up

Is the endpoint reachable? An unauthenticated POST returns 401, not 404:
bash
curl -s -o /dev/null -w "%{http_code}\n" -X POST \ -H "Content-Type: application/json" --data '{}' \ https://<your-ants-host>/api/public/otel/v1/logs
A 404 means the host or path is wrong. The base path is /api/public/otel.
401 errors? Check the Base64 pk:sk value and that the keys belong to the right project.
Everything shows as Claude Code? A stale OTEL_SERVICE_NAME=claude-code is set in the shell. Set it per tool or use a fresh shell.

Codex creates many extra traces

Set trace_exporter = "none" and keep only the logs exporter with the /v1/logs path.

Enterprise MDM rollout

Shadow AI deployment

Integrations overview

Coding-Agent Telemetry

Two ways to set it up

How it works

Authentication

Automatic setup with the enrollment script

Turn on content capture

User attribution

Manual setup

Other tools

What each tool can export

Environment variable reference

Verify it's working

Confirm the configuration

Generate activity

Check ANTS Platform

Troubleshooting

Nothing shows up

Codex creates many extra traces

Related