Tags

Tags are string labels that help you categorize and filter traces. You can apply multiple tags to a single trace or observation, and XeroML automatically aggregates tags from all observations up to the trace level.

Common uses for tags: feature area, endpoint name, experiment name, model variant, or any categorical label that’s useful for filtering.

Tag Limits

• Maximum 200 characters per tag
• Tags exceeding 200 characters are silently dropped
• No limit on number of tags per trace

Adding Tags

Use propagate_attributes() to attach tags to all observations in a context:

Python SDK

from xeroml import observe, propagate_attributes

@observe()
def handle_request(query: str) -> str:
    with propagate_attributes(tags=["feature:search", "model:gpt-4o"]):
        result = run_search_pipeline(query)
        return result

Tags can also be set directly when creating an observation:

from xeroml import get_client

xeroml = get_client()

with xeroml.start_as_current_observation(
    name="my-span",
    type="span",
    tags=["experiment:v2", "region:us-east"]
) as obs:
    result = process()

TypeScript SDK

import { startActiveObservation, propagateAttributes } from "@xeroml/tracing";

await startActiveObservation({ name: "handle-request" }, async () => {
  await propagateAttributes(
    { tags: ["feature:search", "model:gpt-4o"] },
    async () => {
      await runSearchPipeline(query);
    }
  );
});

OpenAI (Python)

from xeroml.openai import openai
from xeroml import propagate_attributes

with propagate_attributes(tags=["endpoint:chat", "version:v2"]):
    response = openai.chat.completions.create(
        model="gpt-4o",
        messages=[{"role": "user", "content": query}]
    )

LangChain (Python)

from xeroml import CallbackHandler, propagate_attributes
from langchain_openai import ChatOpenAI

with propagate_attributes(tags=["pipeline:rag", "retriever:pinecone"]):
    handler = CallbackHandler()
    llm = ChatOpenAI(callbacks=[handler])
    response = llm.invoke(query)

Automatic Aggregation

When tags are applied to individual observations within a trace, XeroML automatically merges all unique tags onto the trace object. This means you can filter traces by tag even when the tag was only applied to a child observation.

Note

Propagation note: Tags set via propagate_attributes() apply to all observations created within that context scope. Tags set directly on an observation apply only to that observation (and propagate up to the parent trace).

Filtering in the UI

Tags appear as filter options in the Traces list view. Select one or more tags to narrow down traces for investigation or dataset creation.