Pricing
Cached answers are free: why repeated AI agent reads cost nothing
Why the cheapest call is the one we never make, and why we bill accordingly.
Most agent traffic is repetition. The same client list, the same open tickets, the same invoice status - asked again and again, by agents that cannot remember each other's answers. Metering every one of those repeats is easy revenue, which is exactly why everyone does it.
What a cache hit costs: nothing
We do the opposite. When actionlake can answer from its cache - and the cache is shared across your tenant, so five agents asking the same question cost one upstream call - the answer costs you nothing: no fee from us, and no call against your vendor's rate limit. The response says AL-Cache: hit on its face, and you can verify it against your vendor's own usage page.
Fair use applies, and it is in the price book too: up to 200 cache hits for every billed call, then US$0.02 per 1,000 hits past that. Writes are never cached and never retried.
Why it is in the terms
This is not a promotion; it is written into the terms. It survives because our incentives are aligned with yours: we make money when agents do new work in your systems, not when they ask the same question twice. A meter that charges for repetition is a meter that quietly punishes agents for being thorough.
Failures are free too
The corollary matters too: anything that fails through no fault of yours - a vendor outage, a timeout, a rate-limit refusal - is also free. You pay for definitive answers. Everything else is our problem, which is what an infrastructure company is for.
The full meter, with every free item, is on the pricing page.
Ask one question across two systems.
100,000 free calls a month. No card. Hard caps on every key. Cached answers are free.
