Helicone

What it is

Observability through a proxy.

Helicone routes your model calls through its proxy: you change the base URL, and every request flows through Helicone on the way to the provider. From that vantage point it gives you logging, cost and latency dashboards, caching, and rate limiting, all with near-zero code change. There is no SDK to weave through your call sites, just a different endpoint and an auth header.

The one job

Instant observability from one line.

Change one thing, the base URL, and you have logging and cost and latency dashboards immediately. It is the lowest-friction way to start seeing your traffic.

Reach for it when

• You want logging and cost/latency dashboards fast with minimal code.
• You want proxy-style caching or rate limits without building them yourself.
• You need a quick win: visibility on day one, not after an instrumentation project.
• You are early and want to defer deeper tooling until the shape is clearer.

Skip it when

• You cannot route through a proxy for governance or data-residency reasons.
• You need deep eval workflows: Langfuse or Braintrust go further.
• You need SDK-level span control over each step: reach for OpenTelemetry.

A minimal look

One line: the base URL.

import OpenAI from "openai";

// the ONLY change: point the client at Helicone + add an auth header
const openai = new OpenAI({
  baseURL: "https://oai.helicone.ai/v1",
  defaultHeaders: { "Helicone-Auth": `Bearer ${process.env.HELICONE_KEY}` },
});

const res = await openai.chat.completions.create({
  model: "gpt-4o", messages,
});                                       // logged, priced, cacheable

The principle it teaches

Where it fits your labs

This is a Week 5 tool. When you want cost and latency numbers on your Lab 5 agent before investing in full tracing, Helicone gets you a dashboard in one edit.