Getting set up, and the command you do not have

The setup lessons in the official course come in two flavours, Python and TypeScript, and they converge on the same place: a CLI, a config file, and a local server you can hit. You are the TypeScript one, so this lesson notes where the two paths differ rather than pretending they are identical.

The four commands you have

langgraph dev runs the server locally without Docker, on port 2024. This is the fast loop. No image build, no container runtime, nothing to pull. On a bad connection it is realistically the only one you can use at all.

langgraph build builds the image. The -t tag flag is required, so a snippet showing a bare langgraph build is wrong on its face.

langgraph up runs the containerised stack, on port 8123. Closer to production shape, needs Docker.

langgraph dockerfile emits a Dockerfile, for when you need to customise the build beyond what dockerfile_lines in the config can express.

The one you do not have

There is a fifth command, langgraph deploy. It is Python-only and in beta.

Not “exists but discouraged”. Not “exists with fewer flags”. It is simply not there for you. Practically, your shipping path as a TypeScript developer goes through the platform. On the exam, if you half-remember a tutorial where somebody typed langgraph deploy and shipped, you were reading Python beta documentation, and building an answer on that memory loses a mark.

Where the config diverges

The same split shows up in langgraph.json. JavaScript and TypeScript configs omit dependencies, python_version, the pip_* keys, and keep_pkg_tools. Your dependencies live in package.json, so the file has nothing to declare. node_version is what you get instead.

This is a real question shape: “which of the following is not available in a JavaScript project”. The answer always comes out of that same five-key Python set, so if you learn one thing here, learn which side of the line each key sits on.

The honest loop

Write the graph and declare it in graphs. Iterate with langgraph dev on 2024. When you want the real artifact, langgraph build -t something, then langgraph up on 8123 to run it the way it will actually be served. Reach for langgraph dockerfile only when the standard build cannot express what you need. Then ship it through the platform, because the CLI shortcut for step five is Python beta and you are not writing Python.

langgraph dev
langgraph build -t my-agent
langgraph up
langgraph dockerfile

{
  "graphs": { "agent": "./src/agent.ts:graph" },
  "env": ".env",
  "dependencies": ["."],
  "store": {}
}