Integrate

@leo-logs/trace does not call the model. It names one run of a pipeline and one run of each agent in it. You put those names on the OpenRouter request you were already going to send. The log page reads them back from the broadcast. The function reference is @leo-logs/trace.

One task, two agents

support-reply answers a customer whose package never arrived. track finds the shipment. answer writes the reply. Both agents belong to the same run, so the session is one cost, one duration, and one failure count.

import { OpenRouter } from '@openrouter/sdk';
import { createPipeline } from '@leo-logs/trace';

const pipeline = createPipeline('support-reply');
const track = pipeline.agent('track');
const answer = pipeline.agent('answer');

const client = new OpenRouter({ apiKey });

await client.chat.send({
  chatRequest: {
    model: 'z-ai/glm-5.3',
    messages: [{ role: 'user', content: 'Where is the package?' }],
    sessionId: track.sessionId(),
    trace: { traceId: track.traceId() },
  },
});

await client.chat.send({
  chatRequest: {
    model: 'z-ai/glm-5.3',
    messages: [{ role: 'user', content: 'My package never arrived. Can you help?' }],
    sessionId: answer.sessionId(),
    trace: { traceId: answer.traceId() },
  },
});

sessionId on the request is the pipeline session. It is the value Leo Logs reads back, so use agent.sessionId() there even if the app was already sending its own session id. Every agent created from that pipeline returns the same session id, so both calls land in one session. trace.traceId is that agent’s execution. Reuse the same agent object, and the same traceId(), on every model call that belongs to that execution. pipeline.agent('answer') a second time starts a second execution.

@openrouter/agent takes the same two fields on callModel: sessionId and trace.

Ids

Each id is chosen once and then stays fixed for that object.

Object	Method	Shape
Pipeline session	pipeline.sessionId()	pipeline:{name}-execution:{uuid}
Agent in a pipeline	agent.traceId()	{pipeline session}_agent:{name}-execution:{uuid}
Agent in a pipeline	agent.sessionId()	The pipeline session
Agent with no pipeline	createAgent(name).traceId()	agent:{name}-execution:{uuid}
Agent with no pipeline	createAgent(name).sessionId()	null

A name may contain hyphens. It cannot contain _agent:.

An agent with no pipeline is its own row on the Agents page. Leave sessionId off the request, and put traceId() on trace.traceId.

examples/calls.ts is a longer task, order-desk, with three agents: lookup, policy, and reply.