nice to see all the crewai examples, thanks. One thing that I like to see is a breakdown of the memory vs context and what exactly do agents store in the context. If have an agent that pulls content from a database, so I pass a page_id to the inputs dict. When the agent finishes, it's extracted a document, but where is that stored, how is it properly referenced and accessed by agents further downstream? Do I specify a key name in the expected_output? ie., store the retrieved document in the context with the key 'database_page'...or something like that? The other frustrating thing is that all the crewai RAG tools point to file paths or URLs etc. So you can't just point to a chunk in context, you have to write out the task output to file and then instantiate the tool... what are your thoughts on the blackbox nature of the context and memory?
ps. is it possible when creating word docs to specify any formatting? Basic things like specifying fonts and adjusting typography or even applying a template. I've been using markdown for everything because its so portable, but it would be nice at the end if there was a straightforward way to dump the crew results into a docx file and make it look nice. Cheers!
anything on fine tuning llms using openvino for cpu purposes pls. it helps with cpu usage and low memory use
nice to see all the crewai examples, thanks. One thing that I like to see is a breakdown of the memory vs context and what exactly do agents store in the context. If have an agent that pulls content from a database, so I pass a page_id to the inputs dict. When the agent finishes, it's extracted a document, but where is that stored, how is it properly referenced and accessed by agents further downstream? Do I specify a key name in the expected_output? ie., store the retrieved document in the context with the key 'database_page'...or something like that? The other frustrating thing is that all the crewai RAG tools point to file paths or URLs etc. So you can't just point to a chunk in context, you have to write out the task output to file and then instantiate the tool... what are your thoughts on the blackbox nature of the context and memory?
ps. is it possible when creating word docs to specify any formatting? Basic things like specifying fonts and adjusting typography or even applying a template. I've been using markdown for everything because its so portable, but it would be nice at the end if there was a straightforward way to dump the crew results into a docx file and make it look nice. Cheers!
Hi, pls tell us how to deploy this on docker or any cloud, u can remove the 2 factor authentication in editing....
What is the best framework for AI Agents, if CrewAI is not the best. What do you recommend?
LangGraph