Sokudo
Matching Rules
A hit is usually prefix-exact
Caching works from the start of the request. Stable content belongs first; volatile content belongs last.
Likely to hit
✓ Same model and compatible settings
✓ Identical system prompt and tool definitions
✓ Same document bytes and ordering
✓ Stable content placed at the beginning
✓ Request arrives within cache lifetime
Breaks or shortens the hit
× Timestamp or request ID near the front
× Reordered tools, JSON keys, or documents
× Whitespace or formatting changes
× Model/version or tenant boundary changes
× Cache eviction or expiration