How I wire AI into a build
A model by itself is just a box that answers. It cannot see your website, your files or your rules until something wires it in. That wiring is this group. These are the kits that carry a question from your website to the model and the answer back, the feature that makes the model point to the exact source it used instead of asking you to trust it, and the trick that lets it reuse work on information it has already seen so the same question does not cost full price to answer twice. You see the result in builds like "Get straight answers about the rules you follow", where every answer comes with the source it came from, and in a website helper that fields the same questions all day without the bill climbing each time.
Anthropic SDK
The official toolkit for talking to Anthropic's Claude models. In plain terms, it is the wiring that carries a question from a build to Claude and brings the answer back reliably and in the right shape. It is a plumbing tool, invisible to your customers, and it is what puts Claude to work inside the builds.
Vercel AI SDK
A toolkit for wiring AI models into a website so answers appear smoothly, a word at a time, the way people now expect. It is what connects the model to a helper living on your site, so a reply streams in rather than making the customer stare at a blank screen. This is builder plumbing behind a website helper like "Answer the questions you get asked all day".
OpenAI SDK
The official toolkit for talking to OpenAI's GPT models, the counterpart to the Anthropic one. It carries a question from a build to GPT and returns the answer. It is one of the reasons a build can use either model maker without being torn apart and rebuilt.
OpenRouter
A single doorway to AI models from several different makers. Instead of wiring up each model separately, a build can reach many through one connection and switch between them as needed. It is what makes "use whichever model does this job best" a setting rather than a rebuild.
Prompt caching
A way of not paying twice for the same work. When a helper answers from the same approved information over and over, the model would normally re-read all of it for every question. Caching lets it hold onto that shared material, so repeat questions come back faster and cost a fraction of the first. It matters most for a website helper fielding the same questions all day.
Citations API
A feature that makes a model show its work. Instead of just giving an answer, it points to the exact passage in your material that the answer came from. That is what lets a customer or an owner check a reply rather than take it on faith, and it is the backbone of "Get straight answers about the rules you follow", where the source matters as much as the answer.
Files API
A way of handing whole documents to a model to work from directly. Rather than copying and pasting text, a build can give the model the actual file and ask questions against it. It is one of the pieces that lets a helper answer from your own documents rather than from whatever it picked up on the internet.