SLA

Service Level Agreement. The contract that says what “we keep it running” actually means, in numbers you can hold someone to. Uptime, how fast we respond when something breaks, how long we have to fix it, what support you get and when.

It exists so neither side is arguing from memory. On AI work it covers the machinery: the model responding, the integrations holding, the service being there. It does not cover whether an answer was any good. That is a knowledge problem, not an availability one, and it gets solved by controlling what the system is allowed to answer from.