Skip to main content
Trace forwarding sends the spans of your automations and AI agents to an OpenTelemetry Protocol (OTLP) endpoint that you control. You set it up in two places: add the endpoint once for your organization, then choose which environments send to it.

Before you start

  • You need the Organization Administrator role to view or change trace forwarding.
  • You need an OTLP traces endpoint that is reachable from the internet, and its credentials. Most tracing backends and the OpenTelemetry Collector provide one.
  • Decide whether to use gRPC or HTTP. Use the protocol your backend documents for OTLP traces.

Add an endpoint

  1. Open Settings icon Organization Settings > Platform > Trace Forwarding.
  2. In General Traces, select Add General Endpoint.
  3. Fill in the connection, authentication, headers, and resource attributes. The fields are described below.
  4. Select Save & Test. Elementum saves the endpoint and sends one test span.
  5. Check the result. Message delivered means your endpoint accepted the test span.
Use a General endpoint for automation spans. GenAI endpoints never receive automation spans. To send AI agent spans to an evaluation platform, see Add a GenAI endpoint. To edit an endpoint, click Edit icon in its row. Editing opens the same dialog.

Connection

Protocol notes HTTP uses the OTLP/HTTP protobuf encoding. A common mistake is to enter only the host for HTTP. The test then fails with an endpoint-not-found error.

Authentication

Choose one Authorization Type: Secret values (tokens, passwords, and custom header values) are write-only. After you save, Elementum shows only whether a value is set. To change one, enter a new value.

Additional headers

Additional Headers are sent with every export, in addition to the authentication header. Use them for non-secret settings your backend needs, for example a dataset or tenant header. Their values stay visible on the settings page, so do not put secrets here. Use Custom Headers authentication for secrets.

Resource attributes

Resource Attributes are key and value pairs that Elementum adds to the OTLP resource of every span sent to this endpoint. Values are sent as strings. Use them to label Elementum spans in your backend. For example: Every span already has service.name set to elementum and, when available, service.version set to the Elementum platform release. A resource attribute you configure with the same key replaces Elementum’s value. For example, setting service.name to elementum-prod changes the service name of every span sent to this endpoint.

Attach an endpoint to an environment

An endpoint receives spans only from the environments you attach it to.
  1. Open Settings icon Organization Settings > Platform > Environments.
  2. On the environment card, click Configure, then Edit Environment.
  3. In Trace Forwarding, select the endpoints that should receive traces from this environment.
  4. Click Save Changes.
Each environment can send to several endpoints, and each endpoint can receive from several environments. Every attached, enabled endpoint receives its own full copy of the spans. Changes can take up to about a minute to take effect.

Add a GenAI endpoint

A GenAI endpoint receives only the spans of AI agents, LLM calls, and tool runs, for an evaluation platform such as W&B Weave. See GenAI tracing for what it receives.
  1. In Settings icon Organization Settings > Platform > Trace Forwarding, go to GenAI Traces and select Add GenAI Endpoint.
  2. Fill in the same fields as for a General endpoint: connection, authentication, additional headers, and resource attributes.
  3. Select Save & Test.
  4. Attach the endpoint to environments, as in Attach an endpoint to an environment.
What differs from a General endpoint: Everything else is the same as for a General endpoint: authentication types, write-only secrets, resource attributes, the test span, attaching to environments, enabling and disabling, and delivery. In an environment’s Trace Forwarding section, General and GenAI endpoints appear in one list, by name. Name GenAI endpoints so you can tell them apart, for example Weave - production. A General endpoint also receives AI agent spans, with the same content, alongside automation spans. You do not need a GenAI endpoint to see AI spans in a general tracing backend.

Enable, disable, or delete an endpoint

  • Disable stops all forwarding to the endpoint, from every environment, without removing its settings. A disabled endpoint is shown struck through in the environment’s Trace Forwarding section.
  • Enable resumes forwarding. Spans recorded while the endpoint was disabled are not sent.
  • Delete removes the endpoint and detaches it from every environment. Click Delete icon in the endpoint row and confirm.

Test the connection

Save & Test, and Test Connection on a saved endpoint, send a single span named elementum.connection-test with your configured headers and resource attributes. The result shows: The test rejects URLs that resolve to private, loopback, or link-local addresses. Your endpoint must be reachable over the public internet. A test that waits longer than 10 seconds for your endpoint fails. The test span is not an automation span. You can ignore or delete it in your backend.

Delivery and failures

Because delivery is best effort, use traces to observe and debug automations. Use the Elementum run history as the record of what ran.

Checklist

  • Endpoint added under General Traces with the full URL. HTTP needs the traces path, for example /v1/traces.
  • Save & Test shows Message delivered, and you found the elementum.connection-test span in your backend.
  • Endpoint attached to the environment on the Environments page.
  • Optional: a deployment.environment resource attribute so you can tell environments apart.
  • Run an automation, wait a few seconds, and search your backend for service.name = elementum.
  • For a GenAI endpoint: send a message to an agent in an attached environment and look for an invoke_agent span.
Next: Continue your trace into Elementum.