Docs/Integrations/Coding Agent Telemetry

OTel Collector

AI coding agents — Claude Code, OpenAI Codex CLI, and GitHub Copilot — can emit OpenTelemetry (OTLP) data that AgenticAnts ingests. This gives you per-tool usage: sessions, token counts, models, and tool calls, attributed to the user and device.

ℹ

Content capture is configurable. The setup below enables prompt/tool capture (OTEL_LOG_USER_PROMPTS, OTEL_LOG_TOOL_DETAILS, Codex log_user_prompt = true) so the trace list shows the real question and tool calls. To capture only usage metadata (counts, models, durations, tool names) and keep prompt/response content out of the telemetry, set those switches to 0 / false (or leave them unset).

How it works

Each tool exports OTLP over HTTP to a single ingest endpoint, authenticated with your project keys. AgenticAnts detects which tool sent the data (from the OTel service.name / scope) and parses it into usage records.

	Value
Endpoint (base)	`https://<your-agenticants-host>/api/public/shadow-ai/otlp`
Protocol	`http/json` (HTTP/protobuf also accepted)
Auth	your project's public + secret API key

ℹ

Replace <your-agenticants-host> with your platform host (e.g. api.agenticants.ai). Get the public and secret API keys from Settings → API Keys.

Authentication

Send your keys as these headers:

code

X-Shadow-Ai-Public-Key: pk-... X-Shadow-Ai-Secret-Key: sk-...

These are equivalent to standard HTTP Basic auth — Authorization: Basic base64(publicKey:secretKey) — so either form works.

Setup

Set the OTel values for each coding agent you use. On macOS you can set them session-wide with launchctl setenv; otherwise add them to your shell profile.

Claude Code reads standard OTEL_* environment variables. Add these to your shell profile (~/.zshrc / ~/.bashrc), or on macOS set them session-wide with launchctl setenv:

bash

export CLAUDE_CODE_ENABLE_TELEMETRY=1 export OTEL_METRICS_EXPORTER=otlp export OTEL_LOGS_EXPORTER=otlp export OTEL_EXPORTER_OTLP_PROTOCOL=http/json export OTEL_EXPORTER_OTLP_ENDPOINT="https://<your-agenticants-host>/api/public/shadow-ai/otlp" export OTEL_EXPORTER_OTLP_HEADERS="X-Shadow-Ai-Public-Key=pk-...,X-Shadow-Ai-Secret-Key=sk-..." # Content capture (prompt + tool details) — required for the full trace view export OTEL_LOG_USER_PROMPTS=1 export OTEL_LOG_TOOL_DETAILS=1 # Flush intervals (ms) — how often logs / metrics are exported export OTEL_LOGS_EXPORT_INTERVAL=5000 export OTEL_METRIC_EXPORT_INTERVAL=10000 # User attribution — ties every trace to a user on the Users page export OTEL_RESOURCE_ATTRIBUTES="user.id=<your-email>,user.account_id=<your-email>"

ℹ

OTEL_LOG_USER_PROMPTS and OTEL_LOG_TOOL_DETAILS capture prompt text and tool inputs so the trace list shows the real question and tool calls. To keep content out of the telemetry, set both to 0 (or leave them unset) — usage metadata (counts, models, durations) is still captured.

Restart your terminal (or the Claude Code session) so the variables are picked up.

Environment variable reference

Tool	Key	Value
Claude Code	`CLAUDE_CODE_ENABLE_TELEMETRY`	`1`
Claude Code	`OTEL_METRICS_EXPORTER`	`otlp`
Claude Code	`OTEL_LOGS_EXPORTER`	`otlp`
Claude Code	`OTEL_EXPORTER_OTLP_PROTOCOL`	`http/json`
Claude Code	`OTEL_EXPORTER_OTLP_ENDPOINT`	`https://<host>/api/public/shadow-ai/otlp`
Claude Code / Copilot	`OTEL_EXPORTER_OTLP_HEADERS`	`X-Shadow-Ai-Public-Key=pk-...,X-Shadow-Ai-Secret-Key=sk-...`
Claude Code	`OTEL_LOG_USER_PROMPTS`	`1` (capture prompt text; `0`/unset to redact)
Claude Code	`OTEL_LOG_TOOL_DETAILS`	`1` (capture tool inputs; `0`/unset to redact)
Claude Code	`OTEL_LOGS_EXPORT_INTERVAL`	`5000` (ms)
Claude Code	`OTEL_METRIC_EXPORT_INTERVAL`	`10000` (ms)
Claude Code / Codex	`OTEL_RESOURCE_ATTRIBUTES`	`user.id=<email>,user.account_id=<email>`
Copilot	`COPILOT_OTEL_ENABLED`	`1`
Copilot	`COPILOT_OTEL_ENDPOINT`	`https://<host>/api/public/shadow-ai/otlp`
Copilot	`OTEL_RESOURCE_ATTRIBUTES`	`service.name=github-copilot-cli,deployment.environment=dev,user.id=<email>,user.account_id=<email>`
Codex	(see `~/.codex/config.toml` `[otel]` block above; `log_user_prompt = true`)	—

OTEL_RESOURCE_ATTRIBUTES ties telemetry to a user (and, when set by the agent, host.id=<device-uuid> ties it to a specific device).

Verify it's working

Confirm the variables are set

bash

# shell session env | grep -E "CLAUDE_CODE_ENABLE_TELEMETRY|OTEL_|COPILOT_OTEL" # macOS session-wide (set via launchctl / the agent) launchctl getenv OTEL_EXPORTER_OTLP_ENDPOINT cat ~/.codex/config.toml # Codex

Generate activity

Run a prompt in the tool (e.g. ask Claude Code a question, or run a Codex/Copilot command).

Check AgenticAnts

Open AgenticAnts — the tool should appear in your usage/traces view with sessions and token counts within a minute or two.

Troubleshooting

Nothing shows up

Endpoint reachable? A quick check — an unauthenticated POST should return 401 (auth required), not 404:
bash
curl -s -o /dev/null -w "%{http_code}\n" -X POST \ -H "Content-Type: application/json" --data '{}' \ https://<your-agenticants-host>/api/public/shadow-ai/otlp/v1/logs
A 404 means the OTLP ingest endpoint isn't available on that host — confirm the host and that the endpoint is deployed.
401 errors? Check the public/secret keys and that they belong to the right project.
Variables not picked up? Restart the terminal/tool. On macOS, launchctl setenv applies to new processes — fully quit and reopen the app.

Codex won't start ("duplicate key")

Codex allows only one [otel] table. Remove any extra [otel*] sections so there's a single managed block.