Portkey observability – Tokenminning

Portkey observability tracks every request through the gateway — model, tokens, cost, latency, and config — for spend attribution and debugging.

This page answers portkey analytics, portkey cost tracking, and portkey observability.

Analytics dashboard

Open Portkey → Analytics:

Metric	Use
Cost by model	Find dominant spend slug
Cost by config	Attribute per feature or environment
Cost by API key	Identify which tool burned the pool
Token volume	Input vs output ratio — context bloat signal

Every gateway request generates a trace with:

Use traces to debug why a frontier model ran when you expected haiku.

Pass metadata in request headers for feature-level slicing:


response = client.chat.completions.create(
    messages=[...],
    metadata={"feature": "code-review", "env": "prod"}
)

Group Analytics by metadata fields in weekly reviews.