Name features, customers and runs

This page follows one example: a support feature that answers a customer’s ticket in two LLM calls, first classifying it, then drafting the reply. Each step works the same whether your calls go through the gateway or straight to your provider, recorded by telemetry: pick your tab.

1. Set up

Gateway
Telemetry

Turn on the x-weflayr-metadata header once: every gateway call, made by any OpenAI client, then carries the values of the scope it is made in.

import weflayr
from openai import OpenAI

weflayr.enable_propagate_metadata_to_gateway()

client = OpenAI(base_url="https://gateway.weflayr.com/v1", api_key="<WEFLAYR_API_KEY>")

const weflayr = require('weflayr');
const OpenAI = require('openai');

weflayr.setup(); // first: it lets the values follow your code into the OpenAI call
weflayr.enablePropagateMetadataToGateway();

const client = new OpenAI({
  baseURL: 'https://gateway.weflayr.com/v1',
  apiKey: process.env.WEFLAYR_API_KEY,
});

Without the SDK, set the same values on each request yourself: see the gateway’s Weflayr fields.

Instrument your provider’s SDK once: every call it makes is then recorded with the values of the scope it is made in. See Telemetry for every provider.

import weflayr
from openai import OpenAI

weflayr.auto_instrument(ai_sdks=weflayr.AiSdk.OPENAI)

client = OpenAI()

const weflayr = require('weflayr');
const OpenAI = require('openai');

await weflayr.autoInstrument({ aiSdks: 'openai' });

const client = new OpenAI();

2. Tag the feature run and name its calls

Wrap the feature in propagate_metadata / propagateMetadata: every LLM call made inside, including nested ones, carries its values, and the scope is one feature run, which AI cost counts and prices. Inside it, set_call_name / setCallName names each call.

feature_name: the product feature making the calls.
customer_name: who the calls are for, to join your AI cost with your revenue.
call_name: one LLM call of the feature. A benchmark is set up per call name.

Gateway
Telemetry

def answer_ticket(ticket, customer):
    with weflayr.propagate_metadata(
        feature_name="support-ticket",
        customer_name=customer,
    ):
        with weflayr.set_call_name("classify_ticket"):
            category = client.responses.create(
                model="claude-opus-5-5",
                input=f"Classify: {ticket}",
            ).output_text
        with weflayr.set_call_name("draft_reply"):
            return client.responses.create(
                model="claude-opus-5-5",
                input=f"Reply to this {category} ticket: {ticket}",
            ).output_text

function answerTicket(ticket, customer) {
  return weflayr.propagateMetadata(
    { featureName: 'support-ticket', customerName: customer },
    async () => {
      const category = await weflayr.setCallName('classify_ticket', async () => {
        const response = await client.responses.create({
          model: 'claude-opus-5-5',
          input: `Classify: ${ticket}`,
        });
        return response.output_text;
      });
      return weflayr.setCallName('draft_reply', async () => {
        const response = await client.responses.create({
          model: 'claude-opus-5-5',
          input: `Reply to this ${category} ticket: ${ticket}`,
        });
        return response.output_text;
      });
    },
  );
}

def answer_ticket(ticket, customer):
    with weflayr.propagate_metadata(
        feature_name="support-ticket",
        customer_name=customer,
    ):
        with weflayr.set_call_name("classify_ticket"):
            category = client.responses.create(
                model="gpt-5.6",
                input=f"Classify: {ticket}",
            ).output_text
        with weflayr.set_call_name("draft_reply"):
            return client.responses.create(
                model="gpt-5.6",
                input=f"Reply to this {category} ticket: {ticket}",
            ).output_text

function answerTicket(ticket, customer) {
  return weflayr.propagateMetadata(
    { featureName: 'support-ticket', customerName: customer },
    async () => {
      const category = await weflayr.setCallName('classify_ticket', async () => {
        const response = await client.responses.create({
          model: 'gpt-5.6',
          input: `Classify: ${ticket}`,
        });
        return response.output_text;
      });
      return weflayr.setCallName('draft_reply', async () => {
        const response = await client.responses.create({
          model: 'gpt-5.6',
          input: `Reply to this ${category} ticket: ${ticket}`,
        });
        return response.output_text;
      });
    },
  );
}

3. Keep one run across processes

Each scope starts a new feature run. When one run spans several processes, for example the ticket classified by your API and the reply drafted by a worker, the two scopes would count as two runs. Create the run’s id once where it starts, send it along with the work, and pass it to every scope of the run.

Gateway
Telemetry

# In your API
feature_run_uuid = weflayr.create_feature_run_uuid()
with weflayr.propagate_metadata(
    feature_name="support-ticket",
    customer_name=customer,
    feature_run_uuid=feature_run_uuid,
):
    with weflayr.set_call_name("classify_ticket"):
        category = client.responses.create(
            model="claude-opus-5-5",
            input=f"Classify: {ticket}",
        ).output_text
queue.send({
    "ticket": ticket,
    "category": category,
    "customer": customer,
    "feature_run_uuid": feature_run_uuid,
})

# In your worker, another process
with weflayr.propagate_metadata(
    feature_name="support-ticket",
    customer_name=job["customer"],
    feature_run_uuid=job["feature_run_uuid"],
):
    with weflayr.set_call_name("draft_reply"):
        client.responses.create(
            model="claude-opus-5-5",
            input=f"Reply to this {job['category']} ticket: {job['ticket']}",
        )

// In your API
const featureRunUuid = weflayr.createFeatureRunUuid();
const category = await weflayr.propagateMetadata(
  { featureName: 'support-ticket', customerName: customer, featureRunUuid },
  () =>
    weflayr.setCallName('classify_ticket', async () => {
      const response = await client.responses.create({
        model: 'claude-opus-5-5',
        input: `Classify: ${ticket}`,
      });
      return response.output_text;
    }),
);
await queue.send({ ticket, category, customer, featureRunUuid });

// In your worker, another process
await weflayr.propagateMetadata(
  { featureName: 'support-ticket', customerName: job.customer, featureRunUuid: job.featureRunUuid },
  () =>
    weflayr.setCallName('draft_reply', () =>
      client.responses.create({
        model: 'claude-opus-5-5',
        input: `Reply to this ${job.category} ticket: ${job.ticket}`,
      }),
    ),
);

# In your API
feature_run_uuid = weflayr.create_feature_run_uuid()
with weflayr.propagate_metadata(
    feature_name="support-ticket",
    customer_name=customer,
    feature_run_uuid=feature_run_uuid,
):
    with weflayr.set_call_name("classify_ticket"):
        category = client.responses.create(
            model="gpt-5.6",
            input=f"Classify: {ticket}",
        ).output_text
queue.send({
    "ticket": ticket,
    "category": category,
    "customer": customer,
    "feature_run_uuid": feature_run_uuid,
})

# In your worker, another process
with weflayr.propagate_metadata(
    feature_name="support-ticket",
    customer_name=job["customer"],
    feature_run_uuid=job["feature_run_uuid"],
):
    with weflayr.set_call_name("draft_reply"):
        client.responses.create(
            model="gpt-5.6",
            input=f"Reply to this {job['category']} ticket: {job['ticket']}",
        )

// In your API
const featureRunUuid = weflayr.createFeatureRunUuid();
const category = await weflayr.propagateMetadata(
  { featureName: 'support-ticket', customerName: customer, featureRunUuid },
  () =>
    weflayr.setCallName('classify_ticket', async () => {
      const response = await client.responses.create({
        model: 'gpt-5.6',
        input: `Classify: ${ticket}`,
      });
      return response.output_text;
    }),
);
await queue.send({ ticket, category, customer, featureRunUuid });

// In your worker, another process
await weflayr.propagateMetadata(
  { featureName: 'support-ticket', customerName: job.customer, featureRunUuid: job.featureRunUuid },
  () =>
    weflayr.setCallName('draft_reply', () =>
      client.responses.create({
        model: 'gpt-5.6',
        input: `Reply to this ${job.category} ticket: ${job.ticket}`,
      }),
    ),
);

Any id you pass must have the shape create_feature_run_uuid / createFeatureRunUuid gives: 32 lowercase hexadecimal characters.

SDKs

Guides

Changelog

Name features, customers and runs

1. Set up

2. Tag the feature run and name its calls

3. Keep one run across processes

​1. Set up

​2. Tag the feature run and name its calls

​3. Keep one run across processes

1. Set up

2. Tag the feature run and name its calls

3. Keep one run across processes