Cohere Monitoring & Observability with OpenTelemetry

What is Cohere Monitoring?

Cohere monitoring gives you real-time visibility into your AI applications by collecting traces using OpenTelemetry. This guide shows you how to instrument a Python application that calls Cohere's Chat, Embed, and Rerank APIs and export its traces to SigNoz, without adding any OpenTelemetry code to your application.

With full Cohere observability in SigNoz, you can trace every Cohere request end to end, correlate LLM spans carrying gen_ai.* attributes such as model name and token usage, set alerts on latency and errors, and continuously improve the reliability of your Cohere applications.

Prerequisites

• A SigNoz Cloud account with an active ingestion key or Self Hosted SigNoz instance
• Python 3.10 or later
• A Cohere API key
• The Cohere Python SDK in your application environment

Monitor Cohere with OpenTelemetry

The instrumentation is no-code: the opentelemetry-instrument launcher discovers the installed opentelemetry-instrumentation-cohere package and instruments the Cohere SDK at process startup, so you don't add any OpenTelemetry code to your application. For more details, refer to the Cohere OpenTelemetry instrumentor.

Step 1: Install the dependencies

Create a virtual environment and install the Cohere SDK, the OpenTelemetry distribution, the OTLP exporter, and the Cohere instrumentor:

python3 -m venv .venv
.venv/bin/python -m pip install --upgrade pip
.venv/bin/python -m pip install \
  cohere \
  opentelemetry-distro \
  opentelemetry-exporter-otlp \
  opentelemetry-instrumentation-cohere

Then install any remaining instrumentation for your application's other dependencies:

.venv/bin/opentelemetry-bootstrap --action=install

The Cohere package must be installed explicitly. The bootstrap command does not add it for you.

Step 2: Configure the OpenTelemetry export

Set these environment variables through your deployment secret manager or a local .env file:

COHERE_API_KEY=<your-cohere-api-key>
OTEL_RESOURCE_ATTRIBUTES=service.name=cohere-otel,service.version=1.0.0
OTEL_EXPORTER_OTLP_ENDPOINT=https://ingest.<region>.signoz.cloud:443
OTEL_EXPORTER_OTLP_HEADERS=signoz-ingestion-key=<your-ingestion-key>
OTEL_EXPORTER_OTLP_PROTOCOL=http/protobuf
OTEL_TRACES_EXPORTER=otlp
OTEL_METRICS_EXPORTER=none
OTEL_LOGS_EXPORTER=none
TRACELOOP_TRACE_CONTENT=false

• <region>: Your SigNoz Cloud region
• <your-ingestion-key>: Your SigNoz ingestion key
• <your-cohere-api-key>: Your Cohere API key.

OTEL_METRICS_EXPORTER=none avoids generic HTTP-client metric volume. Set it to otlp only when those metrics are intentionally required.

TRACELOOP_TRACE_CONTENT=false is recommended in production. The Cohere instrumentor otherwise records prompts, completions, and embeddings in span attributes, which may contain sensitive data.

Step 3: Start your application with instrumentation

Prefix your normal Python command with opentelemetry-instrument:

.venv/bin/opentelemetry-instrument .venv/bin/python app.py

For a web service, preserve its normal command:

.venv/bin/opentelemetry-instrument .venv/bin/uvicorn myapp:app

No CohereInstrumentor().instrument() call is needed in your application code. Once running, each Cohere Chat, Embed, or Rerank request produces LLM-aware spans that carry model, operation, duration, and available token-usage information. Cohere makes API calls only when your code invokes it, so a request must actually run before data appears in SigNoz.

View Cohere Traces in SigNoz

Once configured, your Cohere application automatically emits traces for every Chat, Embed, and Rerank request.

Traces are available in SigNoz under the Traces tab:

When you click on a trace in SigNoz, you'll see a detailed view of the trace, including all associated spans, along with their events and attributes.

Cohere Observability Dashboard

You can also check out our custom Cohere dashboard which provides specialized visualizations for monitoring your Cohere usage. The dashboard includes pre-built charts specifically tailored for LLM usage, along with import instructions to get started quickly.

Additional resources:

• Set up alerts for high latency or error rates
• Learn more about querying traces
• Explore log correlation