Jurisdiction-enforced inferenceAccounts are reviewed per jurisdiction and billed from the first token. No free credits, no promotional tier.Contact us for a quotation

Platform · Caching

Jurisdiction-scoped prefix caching

Prefixes are cached separately per jurisdiction.


Per-jurisdiction
Cache partition
Logged
Hit or miss, per decision
Bounded
Cost uplift of partition

Measured, not estimated

01

Design

How this part works

  1. 01

    Cache partitions

    Prefixes cached in one jurisdiction cannot be reused in another.

  2. 02

    Shared system prompts

    A long system prompt shared by a whole product is cached once per jurisdiction. Prefix reuse reduces cost within each jurisdiction without crossing its boundary.

  3. 03

    Cache hit records

    The selection record states whether a hit was available in the eligible set.

02

Comparison

Boundary-first routing

Global cache

  • Highest hit rate across jurisdictions
  • A warm prefix can pull work into another region

Partitioned cache

  • Hit rate measured per jurisdiction
  • Warm prefixes never cross a boundary
  • Cost optimised inside the policy
03

Elsewhere on the platform

Other platform features