Whetstone.
All courses
langgraphIntermediate

LangSmith Deployment

The Deploy section is 25 percent of the LCAE exam, exactly ten questions out of forty, and it comes from a single official course. This one mirrors it end to end: the vocabulary the whole domain rests on, the langgraph.json contract and the CLI that builds it, the four environments and the arrow between the control plane and the data plane, assistants as configuration objects, connecting a UI and mounting custom routes, the deployment interface you will actually be clicking around in during the exam, the memory module split between checkpointer and store, multi-tenant authentication and authorization with their specificity ladder and their one documented exception, and the deepagents no-code deploy path. The exam is semi-open-book, so this teaches distinctions, architecture, and where to look, not signatures you could paste.

Start the course 19 lessons 312 min
LDCrash course
Curriculum

19 lessons across 7 modules

01Before You Deploy Anything2 lessons · 33 min
02Deploy Your Agent4 lessons · 67 min
03Connecting With Your Agent4 lessons · 64 min
04The Deployment UI1 lessons · 16 min
05Memory3 lessons · 49 min
06Multi-Tenant Auth3 lessons · 52 min
07Deep Agent Deploy2 lessons · 31 min
Capstone

Draw the whole deployment, then defend every box and every arrow

The capstone is a whiteboard, not a repo

You are studying on holiday with barn wifi. You are not provisioning anything. So the capstone is the thing that actually predicts a score on ten exam questions: draw the entire deployment from memory, then defend every box and every arrow on it.

Take a blank page. Draw a control plane and a data plane, and draw the arrow between them pointing the correct way. Inside the data plane, draw one deployment. Inside the deployment, draw two graphs. Hang a default assistant off each graph, then hang a second, differently configured assistant off one of them. Draw a thread. Draw a run connecting an assistant to that thread. Draw Postgres underneath everything and label every single thing it is holding. Draw Redis beside it and label what it holds, which is much less than you expect. Draw the store as a separate box from the checkpointer and label the scope of each. Draw a request arriving from outside, and draw the two handler families it passes through in the right order. Finally, draw a UI and a custom route hanging off the same server, and label what each one needs in order to connect.

Then defend it

For every box and every line, answer out loud, without hedging:

  1. What versions this thing, a revision or an assistant version? What does the other one version instead?
  2. If I rotate a secret, which boxes have to be rebuilt, and which are untouched?
  3. This deployment is in the EU instance and I want it in the US instance. What do I do? The honest answer is short and unhappy.
  4. A request arrives. Which handler family runs first, which runs second, and what are the four things the second family can return?
  5. Three authorization handlers are registered at three different levels. Which one runs?
  6. The server restarts. Which boxes still contain data, which are empty, and which never contained data in the first place?
  7. A user sends a second message while a run is in flight. Name the four strategies and say which one throws away work already done.
  8. What is the difference in scope and lifespan between the two memory boxes, and which one does authorization scope by rewriting a namespace?
  9. My UI is a separate application on a different origin. What does it need in order to talk to this server, and what is it actually speaking?

The bar

You can draw it in under five minutes from memory, and you get all nine defences right first time. Miss one, go back to the lesson it came from and reread only that lesson. This is the whole domain compressed onto a page you can redraw on scratch paper in the exam room before you answer a single question.