Tokenminning with Helicone

Helicone  is an LLM observability platform with an AI Gateway — one OpenAI-compatible endpoint for 100+ models, automatic logging, and 0% markup on provider rates.

Hub for the Helicone cluster. Best fit when you need request-level visibility and throttling before traffic hits providers.

Quick checklist

1. Open Helicone → Requests tab after your first API call — dashboard.
2. Set rate limits per user, feature, or API key — rate limits.
3. Default to gpt-4o-mini or equivalent cheap models.
4. Enable prompt caching proxy for repeat prefixes — caching.
5. Use credits or BYOK — Helicone maintains provider keys for you.

Typical impact: 40–70% savings from rate limits catching runaway loops; 10–30% from prompt cache hits.

How to reduce Helicone spending

1. Measure first

Every request appears in the Helicone Requests tab within seconds — cost, model, tokens, latency.

→ Dashboard

2. Cap spend

Set rate limits (requests per minute, tokens per day) before requests reach the provider.

→ Rate limits

3. Route cheap

Change the model field to switch between 100+ models on the same endpoint. Default to mini/haiku.

4. Cache prefixes

Helicone’s prompt caching proxy stores repeat prefixes — caching.

5. Trim upstream bloat

Reduce context in your agent — Context hygiene.

6. Guardrail in prod

Rate-limit CI and eval workloads separately from production traffic.

→ Rate limits

How Helicone bills

0% markup on provider rates when using Helicone credits. BYOK routes through your own provider keys. See Helicone pricing .

Guides

Connect your stack

import { OpenAI } from "openai";
 
const client = new OpenAI({
  baseURL: "https://ai-gateway.helicone.ai",
  apiKey: process.env.HELICONE_API_KEY,
});
 
const response = await client.chat.completions.create({
  model: "gpt-4o-mini",
  messages: [{ role: "user", content: "Hello" }],
});

Anti-patterns

Related

• Gateways
• Helicone docs