Skip to main content

​
Overview

Cloudflare AI Gateway proxies requests to your LLM providers and can export an OpenTelemetry span for every request it handles. Those spans follow the OpenTelemetry GenAI semantic conventions, so Latitude ingests them directly over OTLP — no SDK or code change in your application. You configure the exporter once in the AI Gateway dashboard, pointing it at Latitude’s OTLP endpoint. Every model call routed through the gateway then shows up in Latitude with its model, token usage, cost, and prompt/response messages.

​
Requirements

  • A Latitude API key
  • A Latitude project slug
  • A Cloudflare AI Gateway with OpenTelemetry export available in its settings

​
Setup

1

Open your gateway's OpenTelemetry settings

In the Cloudflare dashboard, go to AI → AI Gateway, select your gateway, and open Settings → OpenTelemetry.
2

Add a Latitude destination

Click Add Otel Destination and fill in the dialog:Then add two Custom Headers:Replace <LATITUDE_API_KEY> with a Latitude API key (Settings → API Keys) and <PROJECT_SLUG> with your project’s slug.
To keep the API key out of the destination config, store the full Bearer <LATITUDE_API_KEY> value in Cloudflare Secrets Store and reference it from the Authorization header — Cloudflare injects the secret as the entire header value, so it must include the Bearer prefix.
3

Send a request

Route a request through your gateway. The trace appears in Latitude within a few seconds.

​
What gets captured

Latitude resolves the standard GenAI attributes AI Gateway emits:
  • Provider — gen_ai.provider.name
  • Model — gen_ai.request.model
  • Token usage — gen_ai.usage.input_tokens, gen_ai.usage.output_tokens
  • Cost — gen_ai.usage.cost
  • Messages — gen_ai.input.messages / gen_ai.output.messages. AI Gateway puts the raw request body and the upstream provider’s native response in these fields; Latitude unwraps them into a normal conversation view.
AI Gateway reports gen_ai.operation.name as chat for every request, including embeddings. Latitude reclassifies embedding requests to embeddings automatically, so they’re counted and displayed correctly.

​
Linking gateway spans to your app traces

If your application already emits OpenTelemetry traces, you can stitch the gateway span into them by passing trace context on the request:
  • cf-aig-otel-trace-id — a 32-character hex trace ID
  • cf-aig-otel-parent-span-id — a 16-character hex parent span ID
The gateway span is then created as a child of the span you identify.

​
Seeing your traces

Once connected, traces appear automatically in Latitude:
  1. Open your project in the Latitude dashboard
  2. Send a request through your AI Gateway
  3. The request appears with its model, messages, latency, token usage, and cost