https://discord.gg/botpress logo
Streamed responses (like ChatGPT)
# 👀feature-requests
s
Feature request for having chat gpt style chat response.. as it gives feeling of continus stream specifically for long response. @loud-crowd-60573 When will this be implemented?
l
Hey @some-apple-9254 Can you provide an example?
s
When a user asks a question to a ‘Botpress’ chatbot, they receive the answer once it has been fully written. Therefore, they may have to wait anywhere from 15 seconds to much longer (1 minute).
When a user asks ChatGPT a question, they get the answer as ChatGPT writes it. For the user experience with the Botpress system, it is extremely long, even if it is only 15 or 20 seconds. @loud-crowd-60573
l
I think that depends on your workflow! The more work the LLMs have to do, the longer it takes to provide a response! For example I did a basic bot and it responds in less then 2 seconds
s
Yes, I know. But I haven't received an answer to my question. 😢
@loud-crowd-60573
l
Ah - I understand, you want the answer to start being typed out as the final answer is being generated! That is a good suggestion @some-apple-9254 I do not have an exact time for "when" but the team actively reviews all the #1111026806254993459 🫡 This would be amazing implimentation for-sure!
w
@some-apple-9254 that would need a complete review of how Botpress works , since the messages you can send to the user are currently fixed (either a defined text or a variable that is defined above
But indeed that'd be incredible and super satisfying (for the user mainly, to have a Chatgpt experience) , but I don't think we can expect that in a near future
w
And what about others platform whatsapp, telegram, etc they should accept streamed output
s
Okay, but it's extremely frustrating for us and for users, because the system has become low-end. I get feedback at least once a week from customers, etc., asking me when the feature will be available, because the response time is too long.
Given that these are features that have been requested since 2022
w
what llm are you using? if you're using something like claude 4 or gemini 2.5 then streaming probably wouldn't help you a whole lot since the problem is latency (time to first token) and not actual tokens per second
i use openrouter on all my llm generations to try to avoid this but you randomly get like 9+ seconds latency
s
4.1 and 4.1 mini
w
yeah they're pretty slow, streaming would probably help if you're not getting high latency
also maybe try out these llms in blue, they're ranked higher and on the same price range
gpt-5 nano is also about triple the speed