velvet stirrup
Does anyone use anything like prismphp to convert natural language into a SQL query and return the query results as natural language?
In the likes of prismphp, I still can't wrap my head around whether I should give my agent my schema context and let it work out the SQL, or provide it tools that do all the different possible things I think it could/should do.
Is there any risk/problems with supplying it will lots of small tools? How do others approach this?
velvet stirrup
@misty ibex Hey man, sorry for the direct tag. I love the package, its brilliant.
Was just hoping for a slight nudge in the right direction. Should i be creating a tool which calls the llm again to convert natural language to sql? Is that safe to have a llm call within a tool?
I did search for a prismphp discord or something similar but couldn't find one. If there is I'd love to join it
misty ibex
So it totally depends what kind of workflow you are using. You could do it with a single agent and a SQL tool, where the agent has access to the schema, builds the query, then passes the query to the SQL tool which calls a database query. You could also have a parent agent that passes the request to a tool, then that tool calls an agent that has the schema as context, builds the query, pass to the database and then returns a string as the result.
I'm starting a Prism Discord server, access is currently part of my Github sponsor levels for now
velvet stirrup
Excellent thanks for the reply man. This was kinda how I was thinking of doing it. Do you give schema context in a blade file or use a tool to query for it each time?
I didn't realise about the sponser slots. Ill get onto that now
velvet stirrup
Hey @misty ibex i've sponsered, would it be possible to add me to the discord?