Your first trace

The fastest path is wrapping a client you already have.

OpenAI

import OpenAI from "openai";
import { monitorOpenAI } from "@cloudmind-ai/js/openai";

const openai = monitorOpenAI(new OpenAI());

const result = await openai.chat.completions.create({
  model: "gpt-4o",
  messages: [{ role: "user", content: "Write a haiku about observability." }],
});

console.log(result.choices[0].message.content);

Run it. Within a few seconds the trace appears in your workspace with the prompt, the completion, token counts, cost and latency.

Anthropic

import Anthropic from "@anthropic-ai/sdk";
import { monitorAnthropic } from "@cloudmind-ai/js/anthropic";

const anthropic = monitorAnthropic(new Anthropic());

Anything else

Wrap the operation directly. This works for any model, including self-hosted ones.

import { cloudmind } from "@cloudmind-ai/js";

await cloudmind.trace({ name: "classify-intent" }, async (span) => {
  const output = await myCustomModel(input);
  span.record({ input, output, model: "internal-classifier-v3" });
  return output;
});

Adding metadata

Metadata is what makes traces findable later. Attach anything you might want to filter by.

await cloudmind.trace(
  {
    name: "support-summary",
    userId: user.id,
    metadata: { plan: user.plan, tenant: user.orgId, feature: "inbox-summary" },
  },
  async () => summarizeTicket(ticket),
);

Flushing before exit

In short-lived processes such as serverless functions or scripts, flush before the process ends or the final batch is lost.

await cloudmind.flush();

Next: core concepts.