Sokudo
Observability
Track reuse—not just requests
Core ratios
Token hit rate
cached input tokens
───────────────
total input tokens
───────────────
total input tokens
Request hit rate
requests with hit
────────────────
eligible requests
────────────────
eligible requests
Also watchTTFT · input cost · eviction rate · prefix length
Example token hit rate
72%
p95 time to first token
−38%
Cached input share by route