Grok Build OpenTelemetry Monitoring

What is Grok Build Observability?

Grok Build is xAI's terminal coding agent. It ships its own OpenTelemetry exporter, so instrumenting it is a matter of configuration: there is no library to install and no collector to run. Once enabled, it emits metrics covering token usage, turns, tool calls, errors, and startup timings, plus a stream of structured events for each session.

With full Grok Build observability in SigNoz, you can see how much your team is spending in tokens and on which models, which tools the agent actually reaches for, how often turns fail and why, and where startup time is going.

Unlike most integrations in this section, Grok Build has no traces exporter. Everything arrives as metrics under the ai.xai.grok_code meter scope and as OTLP log records. There is no trace waterfall to explore, so the dashboard is the primary view.

Prerequisites

SigNoz setup (choose one):
- SigNoz Cloud account with an active ingestion key
- Self-hosted SigNoz instance
Grok Build installed. See the xAI installation guide
A SuperGrok subscription or an XAI_API_KEY

Monitor Grok Build with OpenTelemetry

Grok Build's telemetry is a double opt-in. The GROK_EXTERNAL_OTEL master switch and at least one exporter must both be set, since either alone emits nothing.

Step 1: Launch Grok Build with telemetry enabled

export GROK_EXTERNAL_OTEL=1
export OTEL_METRICS_EXPORTER=otlp
export OTEL_LOGS_EXPORTER=otlp
export OTEL_EXPORTER_OTLP_PROTOCOL=http/protobuf
export OTEL_EXPORTER_OTLP_ENDPOINT="https://ingest.<region>.signoz.cloud:443"
export OTEL_EXPORTER_OTLP_HEADERS="signoz-ingestion-key=<your-ingestion-key>"
export OTEL_METRIC_EXPORT_INTERVAL=10000
 
grok

Verify these values:

<region>: Your SigNoz Cloud region.
<your-ingestion-key>: Your SigNoz ingestion key.

Grok Build 1.0.3 lists gRPC as a supported protocol, but selecting it fails at export time with gRPC protocol is not compatible with HTTP transport. The telemetry stream still reports itself as active and nothing appears in the terminal, so the failure is silent and no data reaches SigNoz. Run grok --debug and look for external otel: lines if nothing arrives.

Grok Build deliberately has no headers key in its config file, so credentials never touch disk. Every non-secret setting can live in ~/.grok/config.toml under [telemetry] as otel_* keys, but the ingestion key has to come from OTEL_EXPORTER_OTLP_HEADERS.

Step 2: Generate some telemetry

Run a few prompts that make the agent read files or run commands, so tool and token metrics have something to report.

grok -p "List the files in this directory and summarise what this project does."

The default metric export interval is 60 seconds, which is why the snippet above lowers it to 10 seconds while you verify the pipeline. Events are exported every 5 seconds by default.

Most steps are identical. To adapt this guide, update the endpoint and remove the ingestion key header as shown in Cloud → Self-Hosted.

View Grok Build Metrics in SigNoz

Metrics are Grok Build's primary signal. Open the Metrics Explorer, search for grok_code.token.usage, and run the query to confirm data is arriving.

Grok Build token usage metric charted in the SigNoz Metrics Explorer — *grok_code.token.usage charted in the Metrics Explorer*

Every metric lives under the ai.xai.grok_code meter scope. All the counters are monotonic delta sums, so aggregate them with increase to get totals rather than rate.

Metric	Unit	Key attributes
`grok_code.session.count`	`{session}`	base attributes only
`grok_code.turn.count`	`{turn}`	`outcome` (`completed`, `cancelled`, `error`), `model`
`grok_code.token.usage`	`{token}`	`type` (`input`, `output`, `reasoning`, `cache_read`), `model`
`grok_code.tool.usage`	`{call}`	`tool_name`, `outcome`
`grok_code.tool.decision`	`{decision}`	`tool_name`, `decision`, `access_kind`, `permission_mode`
`grok_code.error.count`	`{error}`	`error_category`, `model`
`grok_code.startup.total`	`ms`	`outcome`, `auth_mode`
`grok_code.startup.phase_duration`	`ms`	`phase`, `outcome`, `auth_mode`

session.id is attached to metrics by default. On a large rollout that can be a lot of series, so set OTEL_METRICS_INCLUDE_SESSION_ID=false to drop it. OTEL_METRICS_INCLUDE_ACCOUNT_UUID behaves the same way, while OTEL_METRICS_INCLUDE_VERSION and OTEL_METRICS_INCLUDE_ENTRYPOINT are off by default and can be turned on.

View Grok Build Logs in SigNoz

Open the Logs Explorer and filter with service.name = 'grok-cli' to see the event stream. Each event carries its detail in attributes rather than in the log body, so open a record to inspect it.

A Grok Build event in SigNoz showing its attributes — *Grok Build events carry their detail in attributes such as duration_ms, tool_call_count, and cancellation_category*

Useful attributes include model, input_tokens, output_tokens, cache_read_tokens, reasoning_tokens, tool_name, outcome, and duration_ms. Every event also carries session.id and prompt.id, so you can group everything produced while handling a single prompt.

Grok Build Observability Dashboard

The Grok Build dashboard gives you token usage by type and model, tool activity, error categories, and startup latency out of the box.

Grok Build dashboard in SigNoz — *The Grok Build dashboard template*

Troubleshooting Grok Build Observability

No data in SigNoz

Check that both GROK_EXTERNAL_OTEL=1 and at least one exporter are set. Either on its own emits nothing. Then run grok --debug and search the log under ~/.grok/debug/ for external otel: lines, which report whether the stream resolved its configuration and whether exports are succeeding.

Metrics are missing but events arrive

The two signals have different export intervals. Metrics default to 60 seconds while events default to 5, so metrics can look absent for a full minute after startup. Set OTEL_METRIC_EXPORT_INTERVAL=10000 while you verify.

Nothing arrives for the first few seconds

Grok Build holds telemetry closed at startup until it has fetched fleet policy from xAI, since that policy can disable the stream. The wait is bounded and emission begins within 30 seconds regardless.

Export errors mentioning gRPC

Set OTEL_EXPORTER_OTLP_PROTOCOL=http/protobuf. gRPC is documented as supported but does not work in version 1.0.3.

Instrument the other AI coding agents you run, using the same OpenTelemetry pipeline:

Monitor OpenAI Codex with OpenTelemetry - track Codex CLI sessions, token spend, and command-level traces
Claude Code monitoring with OpenTelemetry - track token usage, costs, and tool decisions
OpenCode observability with OpenTelemetry - trace OpenCode sessions, tool calls, and per-message cost

Browse all LLM observability integrations to instrument the rest of your stack.

Additional resources:

Grok Build Observability & Monitoring with OpenTelemetry