PLATFORMS AND MANAGED SERVICES

HOSTED AGENT PLATFORMS AND RUNTIMES A runtime is the environment in which a program actually runs, including the computing power, the memory, and the supporting software. A hosted agent platform is a service that provides a ready-made runtime for agents, along with extras such as sandboxing, deployment, scaling, and sometimes built-in tools. A sandbox, as you learned in Part Nine, is an isolated space where an agent's actions cannot damage the wider system. Running your own safe sandboxes is hard work, so many platforms sell it as a feature. Deployment means putting your agent into a place where real users can reach it, a subject covered fully in Chapter 58. Scaling means handling more users by using more computing power automatically. What do these platforms offer? They can offer managed sandboxes for running tools and code. They can offer long-running execution, so an agent task that takes minutes or hours is not cut off. They can offer built-in storage for state and memory, as in Parts Six and Eleven. They can offer authentication and permission controls, as in Part Nine. What should you watch for? Portability: how hard is it to move your agent elsewhere if you need to? Data handling: where do your prompts, files, and results go, and who can see them? Pricing: some services charge for compute time, some for the amount of work done, and some for a flat plan, and the cost model can change as your usage grows, a topic you studied in Chapter 53. Limits: every platform has maximum run times, memory sizes, and request rates. Reliability: if the platform has an outage, so does your agent, which links to the reliability ideas of Chapter 56.