For agencies and studios that build and run chatbots for clients: support bots, companion and character products, tutoring bots. One Agency subscription gives you a key per client, a spend cap per key, and a printable report you can hand each client at the end of the month.
The whole product is those three verbs, applied per client.
Each client bot runs on its own API key, and the dashboard breaks estimated input spend down by key, by model, and by day. No more reconstructing who spent what from one blended provider invoice at month end.
Give each client key a hard spend limit that matches the budget you agreed with that client. No single bot can blow the budget, and a runaway loop on one client's product stays that client's problem, not your whole roster's.
60-75% fewer input tokens on long prose conversations, measured in our benchmarks. Support sessions, companion chat, and tutoring dialogs are exactly the workload compression was built for.
Every client key gets a monthly report you can print or export: what the bot spent, what compression saved, and how much traffic it handled. It is the artifact that turns "we manage your AI costs" from a claim into a page you hand over.
That is the white-glove part of the pitch. Your agency runs the engagement and presents the numbers; PromptCrunch appears only in the report footer.
A hypothetical agency running 8 client bots. Your roster will differ; the arithmetic won't.
This is arithmetic on a stated hypothetical, not a promise. The mix decides everything: an agency whose bots are mostly code-heavy or heavily cached would save close to nothing, and we would rather you know that before you subscribe. Run your own mix through the savings estimator.
Compression only pays on long prose conversations. The 60-75% figure comes from our own benchmarks on that workload, not from an industry audit, and it applies past roughly 20 turns of conversation history.
Code-heavy bots saved 0-7% in the same benchmarks, and requests using prompt caching pass through completely untouched. We never break a cache to force savings. If a client's bot is a coding agent or a well-cached agentic loop, tell them caching is the right tool and skip us for that bot; the comparison page says the same thing.
The per-key dashboards and caps still work on passthrough bots. You can run the whole roster through PromptCrunch for visibility and budget control, and let compression pay only where it actually applies.
Any of the 20 API keys on the Agency plan. The intended pattern is one key per client bot: each key carries its own spend limit and its own printable report, so per-client accounting stays separated. Technically a key is just a key and nothing forces the mapping, but the dashboard is designed around it.
Agency caps at 20 API keys. If your roster is bigger than that, email [email protected] and we will work out an enterprise arrangement. The self-hosted build, which has unlimited keys, is on the same path.
Only in the report footer. The proxy is invisible to end users: same models, same answers, no PromptCrunch branding anywhere in the bot itself. The printable client report carries a small PromptCrunch line in its footer; everything else on it is your engagement.
Requests that use cache_control pass through completely untouched, breakpoints intact. That bot keeps its caching savings and gets nothing extra from us. We never break a cache to force savings, so count zero compression savings for heavily cached bots. Details on the vs prompt caching page.
$5 free credit, 100 requests/day. No card. Upgrade to Agency when the roster follows.