#completely

1 messages · Page 1 of 1 (latest)

rich frost
#

lets move convo to here

#

"my only concern is that rewriting system prompt during call but retaining the transcripts will probably bump chance of hallucination"

#

clearly you dont want to just wipe the call transcript right

languid dust
#

no at all

rich frost
#

from your testing would something like adding a mid conversation system prompt like "ignore previous system prompts, your new system prompt is [system prompt you want to add]" be ok

languid dust
#

yeah, I was thinking about something like that

#

using dynamic variables to change some parts of it

#

that would work mid call?

rich frost
#

yeah we're adding functionality so that you can add system prompt mid call and also update dynamic variable

languid dust
#

but as of now not even dynamic variables can update the system context?

#

like, any override calls after the first one would have an effect

rich frost
#

dynamic variables can be in system context but they are only passed once at initialisation

#

this is not currently offered

#

to send override calls mid conversation

#

we are adding

#

PR is being reviewed i believe

languid dust
#

That's really cool

#

by any chance, do you think the Pr would be merged and rolled till monday? hahahahah

rich frost
#

not 100% sure its ready yet

#

let me cehck

#

you have something to deliver by mondaty

languid dust
#

yeah 😅

#

it's okay though, I think we could make do with either a single agent and prompt or juggle multiple agents for now

#

using a custom llm server might be tricky, but also a options for me

languid dust
#

Forgot to bring updates here, the dynamic update feature worked fine us!

#

but since we realized it was implemented kinda like a tool call result to inject context, we got around to actually use a custom llm completion endpoint where we could then overrride manually the system prompt on the context window, which worker a little better for our use case