Memories include relevant links ($100)

[josancamon19]

Is your feature request related to a problem? Please describe.
From the memory processing, extract the category, or topics to search relevant links for.
If can't build a query from memory processing result, update the memory processing and include a parameter for "search" or smth like that, where it generates a query or list of queries from the memory.

Use https://serper.dev/

In order to generate the memory processing params json instructions for LLM, use langchain parser.

Example:

class Model(BaseModel):
    requires_context: str = Field(description='Based on the conversation, this tells if context is needed to answer',
                                  default=False)
    topics: List[str] = Field(description='If context is required, the topics to retrieve context from', default=[])
    dates_range: Optional[Tuple[datetime, datetime]] = Field(description='The dates range to retrieve context from', default=())


parser = PydanticOutputParser(pydantic_object=Model)
print(parser.get_format_instructions())

Some help on the query generation for LLM instructions I was using on a previous project:

def generate_google_search_query(user_input: str):
    return simple_prompt_request(f'''You are a Google Search Expert. You task is to convert unstructured user inputs to optimized Google search queries. Example: USER INPUT: 'Best places to visit in Colombia?' OPTIMIZED Google Search Query: 'Top 10 Colombia tourist destinations'.
    Convert the following user query into a optimized Google Search query: "{user_input}"''')

[mdmohsin7]

As suggested, what if we just include a queries field in the summarizeMemory promot itself? And modifying the prompt a bit to generate possible search words, and then pass them to https://serper.dev

Should I go with this approach @josancamon19?

[mdmohsin7]

Okay just tried the above mentioned approach with a random video and I think it works!