Memory in langgraph
- While building graphs, we have two types of memories
- short term memory: thread specific
- long term memory: user specific
-
Refer Here for official docs.
-
Short term memory with checkpointer will add all the previous conversations to the messages, increasing cost and at some point will break due to token length restrictions.
- Deleting Messages (only 5 recent)
- Restrict based on token count.
- Summarize to restrict
Longterm memory using langgraph
- Lets setup postgres on your system.
-
Install docker desktop
- winget
winget install -e --id Docker.DockerDesktop - homebrew
brew install --cask docker
- winget
-
Linux
docker run -d \
--name postgres15 \
-e POSTGRES_USER=postgres \
-e POSTGRES_PASSWORD=password \
-p 5432:5432 \
postgres:15
- Windows
- Linux
docker run -d `
--name postgres15 `
-e POSTGRES_USER=postgres `
-e POSTGRES_PASSWORD=password `
-p 5432:5432 `
postgres:15
-
To connect to postgres lets use Azure DataStudio
-
In the virtual environment add
uv add langgraph-checkpoint-postgres - Refer Here for memory
Langmem
- Langmem is opensource library by langchain for memory based opeartions
