cicora.ai
API specificationRepeated context

Context caching

Separate recurring context so a client can use compatible caching where the chosen model supports it.

Integration reference: choose parameters and available capabilities from your account API settings and the model catalogue for your environment.

Mark stable context

Cache only the repeated portion: system rules, immutable documentation, or a long reference. A user question and changing data need to remain separate.

The cache_control marker is a compatible-contract preference. Check support in the selected model catalogue before exposing the optimization.

Reusable context marker
json
{
  "model": "anthropic/claude-sonnet-5",
  "messages": [{"role": "system", "content": "Stable product context..."}],
  "cache_control": {"type": "ephemeral"}
}

Invalidate deliberately

  • Change the context version when rules or source material change.
  • Do not put secrets or short-lived permissions into a reusable block.
  • Keep a hash or version of a cacheable source in your own trace.

Verify the effect

Measure latency and usage on comparable requests with the same context.

Do not infer cost from one response: caching depends on model, context size, and request lifetime.