Back to docsRecipe

Per-customer LLM cost attribution

Track token consumption and compute cost per API key, customer ID, or project tag — without modifying your inference code.

Why this matters

LLM bills are opaque. A single $4k monthly invoice tells you nothing about which customer drove the spend. Meridian intercepts every request/response pair, extracts token counts from provider headers, and writes per-customer line items to your existing analytics stack.

How it works

  1. Deploy the Meridian sidecar on your inference hosts.
  2. Tag traffic with a header: X-Meridian-Customer.
  3. Meridian reads x-ratelimit-remaining-tokens and usage headers from every provider response.
  4. Cost is computed against your negotiated pricing table and emitted as structured logs or OTLP metrics.

Example output

{
  "customer": "acme-corp",
  "model": "gpt-4o",
  "tokens_in": 12400,
  "tokens_out": 3100,
  "cost_usd": 0.087,
  "timestamp": "2026-01-18T14:22:01Z"
}

Next steps

Read the sidecar deployment guide to get Meridian running in your environment, then configure your pricing table in the dashboard.