Session setup

Org-level defaults, inherited by every use case. About 2 minutes.

Client name ?

Current total LLM spend / month (USD) ?

baseline for the savings math

Vendor / residency constraints ?

hard filter on the route set

Zero-retention agreements required (PII/PHI/confidential)

Open-weight models required (portability / on-prem someday)

Ops appetite ?

usually kills GPU rental

Coding share of LLM work (%) ?

> 30% forks into harness routing

Existing gateway? ?

LiteLLM / Portkey / other in place

Quick wins already in use ?

caching

batch API

Quality measurement today ?

if "vibes", eval-building becomes action item #1

12-month volume growth ?

Current vendors (free text) ?

Catalog: 26 routes, 25 survive these constraints. show the 1 excluded