+++++1 for this one!
GPT’s are useless for 3rd party apps, if they’re stuck on OpenAI UI.
Don’t you guys wanna make more $$$ with token and requests?
Hopefully in the future we can duplicate the GPT as an Assistant (assuming they use the same API, who knows, Assistants don’t even have GPT-4V or Dall-E yet)
But if you are expecting GPTs to be freely available for API, well, don’t hold your breath. You should’ve been building an Assistant and paying for the service.
Not at all. You can attract & introduce customers that are already Plus subscribers to your services using GPTs & Actions. Users can switch between commonly used GPTs for their daily routines instead of having 3 different apps.
It’s free for you to create and distribute a GPT (And you get revenue sharing) on the platform because it’s expected to stay IN the platform and attract plus subscribers to the GPT ecosystem.
Im not sure you got my point.
If Im building bots, that can be beneficial for our company and employees, and I would like to white label them, GPTS are personalized and polished models that I’d love to use inside my company. But, how can I do that, If our employees has to use OpenAI’s GUI? GPTs stuck inside OpenAI platform, has no use for my use case that Im trying to build. I believe I can fine tune a model there and use it, but… Do you understand my point? Would be much easier to just been able to share the GPTs thru APIs, and of course, we’d pay for it as we already do to use tokens.
- for same requirement, I have created custom gpt using interface and its public. Now I want to access it through API. Mostly from Android / IOS apps. Please assume mobile developers while documenting the API reference docs.
What’s up with the implication that we aren’t “paying for the service” when building a custom GPT?
Custom GPTs and the GPT store are not available to non-plus users.
It’s not an implication, you’re also missing the point.
If you want your GPT to be used on your website you need to create an Assistant and pay for it (for other people to use it)
Of course you need to pay a plus membership to create & use GPTs, but that’s not what I was talking about.
This seems to be weirdly difficult for some users to understand, I guess that’s to be expected when you release something like GPTs. It lowers the barrier to access drastically.
+1 i think this is important because the Assistants do not have web browsing capabilities so +1
Assistants CAN browse the web. GPT’s web browsing capabilities are just functions that call external APIs. You can have your Assistant use more APIs than the base GPT, but you need to code the function execution + retrieval.
I set up an Assistant and asked it to summarize the contents of a web page, given a link. It responded by stating that it cannot browse the web. Perhaps this can be accomplished by way of functions?
+1 to being able to call your own GPT. My main use case is I want to be able to send API calls with context (i.e. knowledge in a custom GPT) but I don’t want to have to pass it in the call every time - mainly due to cost implications.
Or is any of that possible via Assistant API?
API isn’t really for one-on-one interactions. There is an API method to disconnect a retrieval file from an assistant, but that should be considered more for building and revising them. You’d not know when a user is asking a question that is relevant to the retrieval documents.
If you really want control, you’d write an API chatbot that gives you as much control as even enabling or disabling past conversation turns as you see fit. Not an assistant.
+1 for the CustomGPT via API
Hi Thanks for sharing your expertise.
its mean Custom GPT can create by using interface and cannot access via API while Assistance API is not Custom GPT ?
AM i Right?
Thanks
You are right.
“GPTs” are only in ChatGPT Plus, the web chatbot with paid monthly subscription.
“Assistants” is an agent-like interface for API that obfuscates interactions you could program directly with AI models.
This is a massive need. It’s a pity that OpenAI has not made GPT accessible via API. Defeats the entire purpose.
Any workaround anyone has been able to discover to browser internet through Assistants API?
Several posts in this thread state that assistants are custom GPTs with API. However, I haven‘t figured out yet how to enable an assistant to call my own API by providing the OpenAPI 3.0 file and some auth infos.
Would be killer to enable users to interact with your app/service by just chatting with you gpt assistant.
Good lord people…
Lets end this confusion:
GPTs are consumer-facing toys. They are not an API, nor can they be connected to an API. It also sounds like none of you have logged into platform.openai.com and created an ASSISTANT…not a GPT.
I hate to burst the bubble, but unless you’re adding actions, you’re not making a “Custom GPT.” If you will all remember, you had/have Custom Instructions that get applied to your queries. GPTs are just custom instructions with a new UI and the rolling all the different ChatGPT plus features in one. Lets be clear though, you haven’t created anything more than what ChatGPT can do with the right prompt.
Now if youre adding some OpenAPI specs into your Actions option, then you have actually made a personal improvement to the model…but say it with me:
GPT’s ARE NOT THE ASSISTANT API. A GPT CANNOT AND WILL NOT CONVERT TO ANYTHING OFF-PLATFORM.
Okay… so… The Assistant API is how we can interface directly with the different GPT models. It is not a simple task. There is an Assistant creation interface, and you can test chatting with it, but again…If you cant code, then you’re going to have a tough time with this, because its not as simple as using chat.Completions.
chat.Completions is stateless, its a one off message. You have to track all your conversations and feed them back to the endpoint every time… Assistant API is stateFUL and you have to manage Threads of Message and Runs.
Please go read the documentation and youll learn a bunch!
Finally someone who knows what he’s talking about. I asked gpt to summarize the previous post in a nicer tone, here you go:
The author kindly clarifies some common misconceptions about GPT models, emphasizing that they are consumer-focused tools rather than APIs and cannot be directly integrated with external APIs. The writer encourages exploring OpenAI’s platform to understand the distinction between creating an ‘Assistant’ versus using a GPT. They note that without adding specific actions, one cannot create a “Custom GPT,” as it essentially involves applying custom instructions to standard GPT queries. The post further explains the significant difference between GPTs and the Assistant API, highlighting that GPT models are not transformable for off-platform use.
The Assistant API, as pointed out, offers a more complex interaction with GPT models, requiring coding skills, especially for managing stateful interactions, unlike the simpler, stateless ‘chat.Completions’ feature. The author concludes with a friendly reminder to read the documentation for a deeper understanding and more insights.