Add support for VectorStoreRetrieverMemory

[davidmigloz]

https://python.langchain.com/docs/modules/memory/how_to/vectorstore_retriever_memory

[davidmigloz]

@dileep9490 here you have some resources that should help you to implement it:

• Docs
• Python code
• TypeScript code
• TypeScript tests

Dependencies:

• VectorStoreRetriever (already implemented)
• BaseMemory(already implemented)

Sample code that we should be able to run when it's implemented:

final embeddings = OpenAIEmbeddings(apiKey: openaiApiKey);
final vectorStore = MemoryVectorStore(embeddings: embeddings);

final memory = VectorStoreRetrieverMemory(retriever: vectorStore.asRetriever());

memory.saveContext(
  inputValues: {
    'foo': 'bar2',
    'bar': 'foo2',
  },
  outputValues: {
    'foo': 'bar',
  },
);

final memories = await memory.loadMemoryVariables();
print(memories);

[dileep9490]

@davidmigloz can you assign me this issue

[dileep9490]

@davidmigloz how can i get the test data? the type script tests are using TensorFlow embeddings

[davidmigloz]

You can use OpenAIEmbeddings instead that it's already implemented.

[dileep9490]

@davidmigloz that requires my OpenAI API key right if so i'm having some issues with it? or is there any data available?

[davidmigloz]

yes, it requires your OpenAI API key.

To add data to the VectorStoreRetrieverMemory you have to call the saveContext method (you can check the example from the previous comment).

Then when you call memory.loadMemoryVariables(values) it should return the relevant Documents from the vector store.

Another option if you don't want to use OpenAI API is to use a local embedding model using https://www.premai.io/#PremApp. You can still use OpenAIEmbeddings wrapper, but specifying client=OpenAIClient.local('http://localhost:8000') (replace 8000 with the port where your PremApp is running the model).