LangGraph setup and checkpointer packages: what ships in the box

TL;DR: pip install langgraph gets the graph runtime and the in-memory checkpointer. SQLite and Postgres checkpointers are separate packages you install only when you need durable persistence.

The fix

1. In-memory (dev, tests): from langgraph.checkpoint.memory import InMemorySaver - ships inside the langgraph package, nothing else to install.
2. SQLite (local workflows): SqliteSaver / AsyncSqliteSaver - install langgraph-checkpoint-sqlite separately.
3. Postgres (production): PostgresSaver / AsyncPostgresSaver - install langgraph-checkpoint-postgres separately.
4. Persistence only works if you pass a checkpointer to compile(). Without one you get MISSING_CHECKPOINTER the moment you use persistence features.
5. Pass a thread_id in the config ({"configurable": {"thread_id": "..."}}) so conversations checkpoint per thread. No thread id, no resumable history.
6. Async graphs need the async checkpointer variants. Mixing a sync saver into an async graph fails.

When to use this

• This covers exactly what the title says: LangGraph setup and checkpointer packages.
• You are setting this up for the first time, or auditing an existing setup.
• You want the key gotchas in one place before you start.

When not to use this

• You are doing a different workflow with LangGraph; these steps are specific to the title above.
• You need the full reference docs; this is the short path, not the manual.

Compatibility

• Not pinned to a specific version; follows current LangGraph behavior.