Platform · Azure OpenAI

Observe AI agents that run on Azure OpenAI.

Your agent calls Azure OpenAI deployments. Trefur captures each invocation as a span on the agent trace — same shape as OpenAI direct, Anthropic, Bedrock, and Gemini — with deployment, region, tokens, cost, and any tool calls.

What you see.

Deployment name and base model — gpt-4o, gpt-4o-mini, o1, gpt-35-turbo.
Region and resource group per invocation, so you can split by environment.
Prompt tokens, completion tokens, USD attributed per Azure pricing.
Stream vs non-stream invocations captured as one completed span.
Tool calls inside Azure OpenAI chat completions — name, args, response.
Rate-limit headers (429), content filter outcomes, deployment-not-found errors.
Retry attempts and back-off windows as child spans on the trace.

Why it matters.

For engineers

Point your agent's OpenTelemetry exporter at Trefur. Existing code keeps working; Azure OpenAI spans appear on the agent trace as they arrive.

For platform teams

Roll up Azure OpenAI usage per deployment, per resource group, per region. One trace shape across Azure, OpenAI direct, Anthropic, and Bedrock so dashboards keep working through vendor swaps.

For business operators

Know which workflows depend on Azure deployments and what they cost. Catch deployment changes that quietly raise cost-per-run.

For SRE

Azure throttling and content-filter blocks show up on the span. Envelope alerts catch retry-rate spikes per agent and route to oncall with the trace attached.

How it works.

Your agent calls Azure OpenAI through the official SDK or the REST endpoints. Send its OpenTelemetry traces to Trefur and each chat completion, embedding, or assistants call becomes an LLM span. Spans use the OpenTelemetry GenAI shape so they line up with calls from any other vendor.

See what your Azure OpenAI agents are actually doing.

Start free. First trace in under five minutes.

Start freeSee all platforms