feat: add database integration with sqlalchemy about lanarky HOT 10 OPEN

fantastic repo. kindly ask is there any update on this thread?

from lanarky.

@ajndkr would you be able to elaborate on this? Is lanarky meant to solely focused on enabling streaming or does it have a broader goal?

from lanarky.

@awtkns yeah sure. So the project is still holds streaming as the key features but from my experience, there are other common functionalities which can be used via lanarky. The broader goal is to provide necessary tooling for deploying without having to build them from scratch.

For example, tracking tokens per user question is something which is not trivial to build but it needs an sql connection. Hence this particular issues.

from lanarky.

user id, conversation id or session id may also be useful for vector store retrieving

from lanarky.

hi! this feature is still part of the roadmap but i'm unable to find time currently. would you like to contribute?

from lanarky.

@ajndkr glad to contribute if have time.

One question, currently the langchain object is created before passing to the LangChainRouter and wrapped in a simple create_langchain_dependency for dependency injection. Actually, langchain already has session support for memory using sqlalchemy, redis or motorhead, but in order to bind with some user session id, the dependency injection should be more flexible.

For example, instead of the simple create_langchain_dependency, maybe it's better for user to pass the dependency, which may use the session_id in input query or body to construct the memory used for langchain object. Do you have any insights on the ideal API for such things?

https://python.langchain.com/docs/integrations/memory/sql_chat_message_history

from lanarky.

@ajndkr In current implementation, the request and response model are dynamically determined by langchain_object with create_request_from_langchain_dependency. What if the constrcution langchain_object needs to parse request first to get something like session_id for db memory? Can u give me some advice on this?

from lanarky.

@npuichigo hi! create_request_from_langchain_dependency as a feature isn't fully mature to handle all use cases. We add new parameters to handle new requirements but in the meantime, you can check this implementation: https://github.com/menloparklab/chatbot-ui/blob/d2d4aa84ebb6351bfacee99429228d091221b60b/backend/app.py#L101-L114

You can directly use the StreamingResponse class and add custom logic in the wrapped function.

from lanarky.

The code linked uses all the features except add_langchain_api_route part of LangchainRouter. So is that also immature since it uses create_request_from_langchain_dependency?

from lanarky.

The code linked uses all the features except add_langchain_api_route part of LangchainRouter. So is that also immature since it uses create_request_from_langchain_dependency?

nope! the linked code is more low level, so you can skip the use of add_langchain_api_route and use the base fastapi route decorator.

from lanarky.