LLM API
A service that gives programmatic access to language models: your system sends a prompt over HTTP and receives the model's response, usually paying per token processed.
An LLM API is the entry point for using language models inside your own systems. Instead of chatting in a UI, your code makes an HTTP request with the prompt (and parameters like model, temperature and token limit) and gets back the generated text. Providers such as OpenAI, Anthropic, Google and Mistral expose their models this way, and billing is usually per token of input and output.
This is what makes it possible to build AI products without training any model: support chatbots, automatic document summarization, message classification, content generation. There are also aggregators/routers (such as OpenRouter and LiteLLM) that unify several providers behind a single interface, useful for comparing cost and quality or having redundancy.
When to use it: whenever the AI needs to run inside an automated flow rather than a chat window. A concrete example: an email SaaS calls an LLM API for each incoming message, requesting a one-line summary and a suggested reply: the user just reviews and sends.