AI agent hosting glossary

BYOK means you supply your own API key for a model provider (e.g. OpenAI, Anthropic) instead of buying inference through a reseller markup.

An AI agent is software that uses an LLM to decide and take actions — calling tools, running skills, or messaging on your behalf — rather than only answering a single question.

A Docker container is an isolated, lightweight environment that packages an application with everything it needs to run, separate from other containers on the same machine.

MCP is an open protocol that lets an AI model or agent connect to external tools and data sources through a standard interface, instead of a one-off custom integration per tool.

A skill is a discrete capability you add to an AI agent — a plugin that lets it perform a specific action, like calling a webhook, querying an API, or reading a file.

A webhook is a URL that one system calls automatically to notify or trigger another system when something happens, instead of that system having to poll for updates.

A model provider is a company that serves LLM inference over an API — OpenAI, Anthropic, and Google are examples — which you access with an API key you generate and control.

A context window is the maximum amount of text (measured in tokens) a model can consider at once, including the prompt, conversation history, and its response.

Self-hosting means running the software yourself on your own server; managed hosting means a provider runs it for you on their infrastructure for a fee.

Inference markup is when a platform charges more per token than the underlying model provider's own rate, effectively reselling AI usage at a margin.