Definition
LLMOps adapts MLOps to applications that use foundation models instead of models a company trains itself. The emphasis shifts from training pipelines to prompts, retrieval, model choice, evaluation and the behaviour of the application in production.
Typical LLMOps tooling includes prompt registries, eval harnesses, gateways that route between models and enforce limits, tracing, and dashboards for quality and spend. AgentOps extends it further for systems that act.