Crowkis vs Momento: your cache shouldn't bill like the thing it's saving you from
Serverless caches meter every operation. A cache that charges per request in front of an API that charges per request is a strange kind of savings.
Momento's pitch is real: a cache with zero servers to manage, billed per operation. For spiky generic workloads that trade can make sense. But think about what an LLM cache is for — eliminating a per-request bill — and the irony of solving it with another per-request bill becomes hard to unsee. Your savings now have a meter on them.
LLM cache traffic is also the worst case for metered pricing: it's your entire query stream. Every user message probes the cache. At scale, the cache line item starts competing with the model line item, and your finance team starts asking why the cost-reduction component has variable cost. Crowkis is flat — free for Community, per-cluster for Enterprise — so the hundredth million lookup costs what the first did: nothing.
A cost-saving layer should have a fixed cost.
Architecture follows the billing model. A metered cloud cache is an exact-match key-value store over the network; semantic understanding, intent thresholds, poisoning defense, and reasoning reuse aren't on the menu, because the menu is GET and SET. Crowkis spends its entire engine on the question metered caches can't afford to ask: should this answer be reused at all?
The bottom line
There's a deeper point about data gravity, too: your cache contains every question your customers ask. Crowkis keeps that corpus on your hardware, offline-licensed, phoning home to no one. Predictable cost, private data, smarter hits — the meter never had a chance.