Skip to main content

Prove a cheaper model is safe. Then keep the saving.

Cost shadow-runs a cheaper model against your real traffic, has an independent judge grade every answer, and only shows you a switch once it is proven safe. You keep every euro it saves.

Find your safe savings
$npm install @botzone/cost-sdk

Metadata only by default, never your prompts. Open-source SDK, EU-hosted. Free tier, no card needed. Source on GitHub →

Anthropic/OpenAI/Gemini
Open-source SDKEU-hosted in DublinMetadata only by defaultContent-free audit trail
cost.botzone.ai / myapp
Live

Today

€4.82

+12% vs yesterday

Last 7 days

€34.91

€18 saved via downgrades

Top route

chat:reply

€18.20 · 52% of spend

Spend, last 24h

EUR · UTC

By route

  • chat:reply18.20
  • summary:thread7.40
  • classifier:intent4.90
  • embed:search2.60
  • other2.20

Preview · seed data

One call. Three providers.

Wrap your client. That's the whole SDK.

Auto-detects Anthropic, OpenAI, and Gemini. Adds zero measurable latency. Tag every call with a route and Cost does the rest.

import Anthropic from "@anthropic-ai/sdk";
import { wrap } from "@botzone/cost-sdk";

const anthropic = wrap(new Anthropic(), {
  apiKey: process.env.COST_API_KEY,
  route: "chat:reply",
});

// Use it exactly like before. Cost attributes every call.
const response = await anthropic.messages.create({
  model: "claude-sonnet-4-7",
  max_tokens: 1024,
  messages: [{ role: "user", content: "Hello" }],
});

How it works

Three steps. Two minutes.

  1. 01

    Install

    Pull the SDK into the service that calls Anthropic, OpenAI, or Gemini.

    npm install @botzone/cost-sdk
  2. 02

    Wrap your clients

    One call. The SDK auto-detects the provider and forwards every request unchanged.

    const ai = wrap(new Anthropic(), { route: "chat:reply" });
  3. 03

    See where the money goes

    Open the dashboard. Spend attributed to routes and features, with verified-safe downgrade suggestions ready to apply.

    cost.botzone.ai/projects/myapp

The three pillars

Prove. Attribute. Cut.

Verified-safe downgrades

When a cheaper model can do the job, Cost shadow-runs it on your real traffic and has an independent model grade every answer. You only see the switch once it is proven safe, and the risky cases stay on the premium model.

Per-route attribution

Tag every call with a route or feature. The dashboard tells you which prompts are burning your budget.

Three SDKs, one ledger

Anthropic, OpenAI, Gemini. TypeScript and Python. One wrap call, every euro accounted for.