LangGraph Platform
A Python file on your laptop is not a product. Platform wraps invoke, stream, and resume so a web app can talk to one durable graph.
Your graph is the kitchen. Platform is the restaurant: tickets (threads), a pass (API), and a window where the chef watches orders (Studio).
Visual Workflows
Start here — scroll inside each diagram frame to explore, then use + / − to zoom up to 200% if needed.
Overview
Scroll inside the frame to explore · use + / − to zoom up to 200%
Scroll inside the frame to explore · use + / − to zoom up to 200%
Local loop with Studio
Scroll inside the frame to explore · use + / − to zoom up to 200%
dev server first. Click a thread. See the same checkpoints you printed in the terminal.
What you deploy
Scroll inside the frame to explore · use + / − to zoom up to 200%
You deploy a compiled graph plus a checkpointer backend. Not a notebook. Not MemorySaver.
Key Takeaways
- 1.LangGraph Platform is how you run the compiled graph as a server: threads, assistants, streaming, HITL over HTTP. langgraph.json points at your graph. langgraph dev is the local server plus Studio.
- 2.Studio is the visual debugger for threads — you watch nodes light up, you resume interrupts. Production is LangSmith deployment (hosted) or your own LangGraph Server. Same thread model you already learned.
- 3.A langgraph.json file names the graph export. langgraph dev runs an in-memory or configured server and opens Studio. Production: attach Postgres, env secrets, and a real thread store.
- 4.LangSmith traces still sit beside this — Platform is the runtime host, LangSmith is observability.
Learn elsewhere
- →Production agents — Phase 21
- →LangSmith — Phase 19
Real Example
Scenario
Same support graph from the last module, served locally. The UI posts to /threads/{id}/runs. You approve a refund in Studio. The thread continues.
What you would do
Do not start Platform on day one. After the last module's graph.py runs in a terminal, wrap it. If Studio is confusing, your graph is confusing — simplify nodes first.
Commands
Commands to Remember
pip install -U "langgraph-cli[inmem]"langgraph dev # local server + Studiolanggraph.json names the graph to servePostgres checkpointer in production, not MemorySaver
Cheat Sheet
Quick recap
quick ref- •Server hosts the graph
- •Studio debugs threads
- •langgraph.json
- •Prod needs Postgres
Common Mistakes
- ✕Deploying MemorySaver
- ✕Skipping thread_id in the client so every click is a new conversation
- ✕Opening Platform before you can explain your own nodes