Skip to main content
Open WebUI supports OpenTelemetry export for its backend services. Point its OTLP exporter at Phoenix to see backend spans for each request — no code changes required.
This page covers Open WebUI’s built-in backend telemetry. It captures service spans such as FastAPI routes, database calls, Redis calls, and outbound HTTP requests. It is not a full LLM conversation tracing integration by itself. See Add richer LLM traces.

​
Prerequisites

  • A self-hosted Open WebUI deployment
  • A running Phoenix instance that Open WebUI can reach

​
Launch Phoenix

The fastest way to get started with Phoenix is by signing up for a free Phoenix Cloud account. If you prefer, you can also self-host Phoenix next to Open WebUI. Note your Phoenix endpoint — for example http://localhost:6006 for a local instance or https://app.phoenix.arize.com/s/<your-space-name> for Phoenix Cloud — and, if authentication is enabled, an API key from the settings page.

​
Configure Open WebUI

Set these environment variables on the Open WebUI service. They switch the exporter to OTLP over HTTP and send spans to Phoenix’s /v1/traces endpoint:
OTEL_RESOURCE_ATTRIBUTES sets the Phoenix project through the openinference.project.name resource attribute. Without it, Open WebUI traces land in the default project.
Open WebUI may attach client.user.email to authenticated spans. This setup sends those spans to Phoenix. If you don’t want email addresses stored, remove this attribute before export with an OpenTelemetry Collector.
With OTEL_OTLP_SPAN_EXPORTER=http, Open WebUI uses OTEL_EXPORTER_OTLP_ENDPOINT as-is, so include the /v1/traces path. Open WebUI defaults to OTLP over gRPC; a self-hosted Phoenix also accepts gRPC on port 4317, but HTTP works for both self-hosted Phoenix and Phoenix Cloud.
If your Phoenix instance has authentication enabled (for example, Phoenix Cloud), point the endpoint at your instance and pass your API key with the standard OpenTelemetry headers variable:
Leave OTEL_BASIC_AUTH_USERNAME and OTEL_BASIC_AUTH_PASSWORD unset. Open WebUI only falls back to OTEL_EXPORTER_OTLP_HEADERS when basic auth isn’t configured.

​
Run Open WebUI in a container

This example runs Open WebUI with tracing pointed at a Phoenix instance on the host:

​
Observe

  1. Use Open WebUI and trigger backend requests, such as signing in or sending a chat message.
  2. Open Phoenix and select the open-webui project.
  3. Look for Open WebUI spans as requests complete.
Open WebUI’s built-in OpenTelemetry instrumentation includes spans for:
  • FastAPI routes
  • SQLAlchemy database queries
  • Redis calls
  • outgoing HTTP requests made with requests, httpx, and aiohttp
  • errors and latency on backend operations

​
Add richer LLM traces

Open WebUI’s built-in telemetry is useful for backend operations, but it doesn’t include the full prompt, response, token, tool, and model metadata you expect from LLM observability. For richer LLM traces, instrument the model-provider path that Open WebUI calls:
  • If Open WebUI sends traffic through an OpenAI-compatible gateway or provider, instrument that gateway or provider path with the matching OpenInference integration.
  • If you maintain a custom Open WebUI Pipeline or filter, emit OpenTelemetry spans that follow OpenInference semantic conventions and export them to Phoenix.
  • If Open WebUI calls a local model server such as vLLM, combine this page with vLLM tracing.

​
Troubleshooting

  • No traces in Phoenix. Confirm ENABLE_OTEL=true, ENABLE_OTEL_TRACES=true, and OTEL_OTLP_SPAN_EXPORTER=http, and that the Open WebUI container can reach the endpoint. The endpoint must end in /v1/traces.
  • Traces land in the default project. Confirm OTEL_RESOURCE_ATTRIBUTES includes openinference.project.name=open-webui.
  • Spans appear but LLM details are limited. Open WebUI’s built-in telemetry is backend telemetry. Add OpenInference instrumentation to the provider, gateway, or custom Pipeline path for richer LLM spans.

​
Resources

Open WebUI OpenTelemetry

OpenInference