One signed Docker image. Every feature compiled in. Free to run. docker pull crowkis/crowkis:latest
← back to the Roost
vs the fieldApril 1, 2026· 3 min read

Crowkis vs AWS Bedrock prompt caching: the cloud's cache serves the cloud

Bedrock's caching cuts repeated-prefix costs inside one cloud's model garden. Your cache strategy deserves a longer horizon than a vendor's feature page.

Bedrock prompt caching follows the provider playbook: repeated prompt prefixes within a short window bill cheaper on supported models. Inside an all-in AWS stack it's a sensible discount to collect. But inventory what it is — input-side, prefix-exact, minutes-lived, model-gated, and bound to one cloud's catalog. Four adjectives and a leash.

In plain words: Provider caches are loyalty cards. Crowkis is your own wallet — it works in every store, and nobody can cancel it.

The strategic cost is the leash. Build your savings around one cloud's caching semantics and you've added gravity to a decision — model choice — that the last two years have proven you'll revisit quarterly. The best model for your workload keeps changing vendors; a cache that can't follow is a sunk cost waiting to be recognized.

model upgrades without the cold start

The upgrade is a workflow, not a leap of faith.

Crowkis is deliberately provider-promiscuous: it fronts Bedrock, OpenAI, Anthropic, Gemini, and your local vLLM with the same engine, persists answers durably on your disk, and — at the Enterprise tier — bridges cached answers across providers, so switching models doesn't cold-start your memory. The cache becomes the stable layer in a stack where everything else churns.

The bottom line

Collect every provider discount; they stack with us happily. Just keep the institution's memory in something the institution owns.