https://discord.gg/botpress logo
Powered by
Tired of the basic "Generate AI Text" card? Lookin...
# šŸ¤help
a
Hey everyone! šŸ‘‹ Is anyone else finding the basic "Generate AI Text" card a bit too... basic? I feel like it constantly gives me the same robotic answers, no matter how much I tweak the prompt. It doesn't feel like there's actual "AI" reasoning behind the responses sometimes. I'm looking for a better way to use Gemini (Google AI) within my workflow. Specifically, I want a card where I can: Provide a clear System Prompt (instructions/personality). Pass the user's input. Crucially: Save the generated output directly into a string variable (like workflow.my_variable) so I can use it later in the flow (or send it to my app) instead of just displaying it immediately. I've seen the Google AI: Generate Content cardā€”for those using it, does it handle complex instructions better than the native text card? And are there any tips for mapping the response to a specific variable so it stays "intelligent" and varies the phrasing?
l
Will get back based on our conversation in the AMA
This was using the open AI generate content card
a
hello @loud-crowd-60573 i came to a conclusion according to the requirements my boss ask of me
i will make a huge post later today to ask the community of botpress experts about their point of view
l
You can always upgrade to plus plan and have direct access with the solution engineers
w
Hey, for the gemini content card, I think how well it handles complex instructions depends on 1. The model 2. The prompt You can def store the result in a variable to use later but best practice here is to run an execute code card to get the content
The "Generate Content" card will return a full on json in the var you choose to store it in. ^Which is why the execute code card i'd say is necessary You could pass that full json to an autonomous node but that'll bloat the context
a
my company is already at plus plan can i ask your engineers for a full review of my roadmap ? @User
l
You can try! Create your thread anyways and other community members can chime in too.
a
ok i will try to put a super explicative text about my problematic and the solution i came to
it will take time for me to write that
@loud-crowd-60573 @white-exabyte-40437 do you guys have the same problem as me ? Iā€™m currently building a V2 agentic architecture and Iā€™m hitting a massive wall with the Generate Content cards (specifically for Gemini/LLMs). The UI is not just counter-intuitive, it seems broken: The "Missing Field" Loop: Even when following the nested structure (Contents > Role > Parts > Type: Text), the actual input field to type the text simply does not appear or remains hidden. I am getting Google AI error: contents are required because there is physically no place to define the message content in the current UI state. The Variable Mode {x} Trap: Because the manual input field is missing/hidden, the natural instinct is to toggle the {x} (variable mode) on the Messages header. But this expects a structured Array, and passing a simple string like {{event.preview}} throws a string type is not allowed error.
w
From what I know the Gemini card needs message history for it to work, this is the array itā€™s expecting in the messages field
I donā€™t know the structure of that array though
U can try using the openai generate card, it will work without a message history ā€” this is what I did to get it working fast
@aloof-napkin-97210 ^
a
i will try generate content from open ai
right now
you are right the open ai generate content works
i have a question : generate content card vs automnous node for the same input with the same prompt as instructions do you think its normal that the answer generated is always the same ?
when i try with an llm outside of botpress its not the case
even if i keep saying the same thing the answer each time is different
closing this topic whatever "Generate content Card" you use all of them are terrible i recommend to customize automnous nodes to work how you want instead
PreviousNext
Powered by