It is hard to customize the implementation of the checkpointer

[HoangNguyen689]

Background

• Langgraph has a mechanism to store graph state information through checkpointer.
• Langgraph has a tutorial about persistent storage here
• Tutorial describes the persistent memory in many ways
  • in-memory
  • customize in Postgres
  • customize in Redis
  • customize in MongoDB

Problem

• Tutorials are inconsistent.
  • Some examples are out of date compared to others.
  • Only the PostgreSQL example is the most complete and up-to-date.
• Langgraph allows users to customize the persistent memory, but there isn't any instruction about how the checkpointer is stored
  • Although there are examples of storing it in some databases, the core concept has yet to be described.
  • For example, which table is needed?

Proposal

• Instead of giving detailed examples, which can become obsolete at any time (because the development speed of Langgraph is fast), we should have the documentation about how the checkpointer is stored, which resource is needed.
  • That will help everyone can easily understand the way checkpointer works and can customize it on their own.
  • That also helps the documentation more readable.
• The previous examples should be kept track of the latest version of the library.

[vbarda]

thanks for the feedback! we're working on improving the docs / developer experience around checkpointers and will post an update soon.