The names that lie
Before any of the architecture, you have to survive the branding. This domain got renamed and the tooling did not follow, so the words on the documentation page and the words in your terminal openly disagree with each other. That disagreement is not a trivia problem. It is a lookup-speed problem, and lookup speed is the resource this exam actually charges you for.
What things are called now
The managed product you deploy to is LangSmith Deployment. The runtime that serves your graph, the process exposing the API endpoints your client hits, is the Agent Server.
Learn those two as a pair, because a surprising number of questions in this domain are secretly asking “which layer are we talking about”, and the layer is named right there in the question stem.
What the tooling still calls itself
Now the fun part. The rename never reached the CLI, the config file, or the licensing.
- The CLI is still
langgraph. - The config file is still
langgraph.json. - The self-hosted license variable is still
LANGGRAPH_CLOUD_LICENSE_KEY.
The docs site trap
The old documentation lived at langchain-ai.github.io/langgraph. It was canonical for years, which means it is still bookmarked, still linked from blog posts, and still the first thing your fingers type.
That site now only serves a redirect. The live reference is docs.langchain.com, which is also one of the two sites you may consult during the exam, alongside smith.langchain.com.
That matters more than it sounds. Semi-open-book means your constraint is not knowledge, it is retrieval. Forty questions, a hundred and twenty minutes, three minutes each. Ninety seconds spent chasing a retired product name through a redirect is half a question’s budget spent on nothing.
The tell you can use on every question in this course
When an answer option uses a name, ask which layer it names: product, runtime, or tooling. Options that quietly mix layers, something like “deploy your Agent Server config to LangGraph Cloud”, are usually built out of exactly this confusion. Once you can see the seam, the distractor stops being camouflaged.
Try it yourself
Product name versus runtime name
Two different layers got named in this domain, and questions love to swap them.
What is the current name of the managed product you deploy to, and what is the name of the runtime that actually serves your graph?
Reveal answer
The product is LangSmith Deployment. The runtime that serves the graph is the Agent Server. The product is where you deploy; the Agent Server is the process running underneath that exposes the API your client calls. The word "LangGraph" survives in the tooling names, the langgraph CLI, langgraph.json, and LANGGRAPH_CLOUD_LICENSE_KEY, but it is not the current name of either the product or the runtime.
Which legacy name is still correct to type
The branding moved on. The tooling did not. Exactly one of these is a name you should still be typing today.
Show answer
Correct answer: B — The config file langgraph.json, which is still the real filename the CLI reads
langgraph.json is genuinely still the filename, alongside the langgraph CLI and the LANGGRAPH_CLOUD_LICENSE_KEY variable. The tooling kept the old name even though the branding did not. The tempting wrong answer is the docs site, because it was the canonical reference for years and is still deep in muscle memory and in half the Stack Overflow answers you will ever read. It now only serves a redirect, so citing it as current is a mistake. The product is LangSmith Deployment and the runtime is the Agent Server.
Where you are allowed to look
The exam is semi-open-book. Two sites are whitelisted, general web search and AI assistants are not. You need the current key set for a config file and you have two minutes.
Show answer
Correct answer: C — docs.langchain.com, the live reference, which is one of the two whitelisted sites
docs.langchain.com is the live reference and it is whitelisted along with smith.langchain.com. Nothing else is. The tempting wrong answer is the legacy docs site, because it genuinely was the right answer for a long time and it will still resolve in your browser rather than throwing an obvious error: it redirects, which costs you the time you were trying to save. The GitHub README is not whitelisted and general search is explicitly out.
Why naming still costs marks when the book is open
It is easy to assume an open book makes vocabulary free. It does the opposite here.
The exam is semi-open-book. Why does knowing the current product and runtime names still matter?
Reveal answer
Because you cannot look something up efficiently if you are searching for a name that no longer exists in the documentation. Searching "LangGraph Cloud" or landing on the redirected legacy site burns minutes you do not have across forty questions in a hundred and twenty. Knowing that the product is LangSmith Deployment and the runtime is the Agent Server means your lookups land first time. It also means a stale name inside an answer option becomes a tell rather than a distraction.