Response Cache

Overview

The gateway can serve a repeated completion from cache instead of re-running the model. This endpoint group lets you see what your tenant currently has cached and throw it away when the underlying facts change.

Every entry is tenant-scoped. You never see, count or purge another tenant's cache.

A typical use: your knowledge base changed, so the cached answers about it are now wrong. Purge the cache and let the next request repopulate it.

Authentication

Authorization: Bearer cm_your_api_key

GET /api/v1/gateway/cache/stats

Counts only live, unexpired entries.

curl https://api.callmissed.com/api/v1/gateway/cache/stats \
  -H "Authorization: Bearer cm_your_api_key"

{
  "entries": 1284,
  "total_hits": 9317,
  "bytes": 4218904,
  "cached_prompt_tokens": 2841002,
  "cached_completion_tokens": 512884
}

The two token totals are the value the cache has returned so far — multiply them by the model's rate to see what you saved.

DELETE /api/v1/gateway/cache

Purges every cached entry for your tenant.

curl -X DELETE https://api.callmissed.com/api/v1/gateway/cache \
  -H "Authorization: Bearer cm_your_api_key"

{ "deleted": 1284 }

Safe but not free: the next request for each purged prompt runs the model again and is billed normally.

DELETE /api/v1/gateway/cache/{cache_key}

Purges one entry.

curl -X DELETE https://api.callmissed.com/api/v1/gateway/cache/3b9f…c1 \
  -H "Authorization: Bearer cm_your_api_key"

{ "deleted": 1 }

A key that is not 64 hex characters returns 422. A well-formed key with no live entry returns 404 Cache entry not found.

Errors

Reading stats and purging never consume credits.