The Routeplane blog
Field notes on sovereign AI gateways, per-request residency, guardrails, and the regulated edge of LLM infrastructure.
TensorZero is archived: where the flywheel goes next
TensorZero articulated the inference flywheel better than anyone, then wound down. A concept-by-concept migration map for its users, with the gaps stated plainly.
7 min read →
Prompt caching changes the economics of context compression
A 44% “saving” from trimming old context can be worth 13% once prompt caching is priced in. The measurement, why it happens, and the eviction strategy that keeps 36%.
7 min read →
What is a sovereign AI gateway?
Deploying in a region tells you where your cluster runs, not where each request’s data went. A sovereign AI gateway closes that gap by enforcing residency per request.
6 min read →
DPDP-compliant LLM inference: keeping data in India
India’s DPDP Act makes "where did this prompt’s data go?" a question with legal weight. Here’s how to answer it per request, without routing everything to one provider.
7 min read →
What is an LLM gateway?
One API in front of every model provider, with routing, fallback, guardrails, caching, and cost control. The what, the why, and where sovereignty fits in.
5 min read →