Anyone know where i can find more info

Anyone know where i can find more info on chatstorage ? Persist to db, recall history load sessions and so forth?

15 comments

Its a pretty new concept -- I wrote it because someone was adding a redis memory module, and I didn't want to maintain two versions of the same sliding-window buffer logic 😅

Docs are here
https://docs.llamaindex.ai/en/stable/module_guides/storing/chat_stores.html#chat-stores

Happy to answer any questions though that are maybe missing from that page

hhansson0728

Ok cool, yeah i read the docs, thought i might have missed some, but basically saving a conversation to json then reopen it as_chat will have a menory of the whole conversation ?

LLogan M

Yea, like you'll need to load the chat store, and pass it to the memory module with the appropriate key set.

Plain Text

chat_memory = ChatMemoryBuffer.from_defaults(
    token_limit=3000,
    chat_store=chat_store,
    chat_store_key="user1",  # <- points to what collection in the chat store to use
)

Then using that memory in a chat engine or agent, it will have access to the past chat history

hhansson0728

I cant get it to work the storage is never created i e setup a redis stack to run against

LLogan M

mmm it works for me?

Plain Text

from llama_index import SimpleDirectoryReader, VectorStoreIndex
from llama_index.storage.chat_store import RedisChatStore
from llama_index.memory import ChatMemoryBuffer

chat_store = RedisChatStore(redis_url="redis://localhost:6379")
memory = ChatMemoryBuffer.from_defaults(chat_store=chat_store, chat_store_key="user123")

documents = SimpleDirectoryReader("./docs/examples/data/paul_graham").load_data()

index = VectorStoreIndex.from_documents(documents)

agent = index.as_chat_engine(memory=memory)

agent.chat("What did the author go growing up?")
agent.chat("Thanks!")

print("agent ", agent.chat_history)
print("-------")
print("chat_store ", chat_store.get_messages("user123"))

LLogan M

Also just running redis-stack on docker

Attachment

hhansson0728

Ahh now i see it i loaded it in the wrong way, thanks logan!! Iam sorry but iam sure i will be back with stupid questions 🙂

hhansson0728

Are there any plans for more stores ?? Postgres ?

LLogan M

I'm not sure how storing a list would work in postgress. A JSON blob maybe?

The interface is fairly simple -- very open to community PRs ❤️

hhansson0728

Ive been working with python for like 1 week 🙂 so not yet i think, i can probably write a wrapper on the simplechat store to store in postgres

LLogan M

If you feel brave, its only one file actually. This is the base class: https://github.com/run-llama/llama_index/blob/main/llama_index/storage/chat_store/base.py

And here's the simple chat store
https://github.com/run-llama/llama_index/blob/main/llama_index/storage/chat_store/simple_chat_store.py

hhansson0728

Yeah i know looked at it already 🙂

LLogan M

ah sweet 🙂

hhansson0728

Iam thinking a chatstore table with a combined pk on sessionid and message id would work the. Just do inserts upserts deletes on the table would work, but i dont have the knowhow to do it yet though

hhansson0728

hmm when running a
"
Stream = agent.stream_chat(prompt)
for response in Stream.response_gen:
responseBuffer += (response or "")
"

the assistans answer is not appended to the chatmemory

Add a reply

Find answers from the community

Anyone know where i can find more info