Integration modes · Updated 2026-07-21

Request metadata

Attach content-free workload labels and trace IDs so Varsten can allocate spend and analyze agent workflows.

What metadata is for

Request metadata tells Varsten which product feature, team, customer, environment, or workflow generated a call. These labels power allocation and task-aware analysis.

OpenAI wrapper example

ts
import { VarstenTrace } from "@varsten/openai";

const trace = new VarstenTrace();

await client.chat.completions.create(body, {
  varsten: trace.metadata({
    feature: "research_agent",
    team: "product",
    taskType: "research.step",
    riskLevel: "medium",
  }),
});

Supported labels include trace, feature, workflow, customer, external user, team, department, environment, task type, risk level, and quality threshold.

Content boundary

Metadata values must be short labels or identifiers. Never put prompts, completions, provider keys, access tokens, customer content, or other secrets in metadata.