Creating Custom GPTs: API Key For Custom Actions | Complete Guide
HTML-код
- Опубликовано: 12 ноя 2023
- This video provides a comprehensive guide on how to create custom GPTs with a focus on integrating Auth Actions API Key for applications in the OpenAI App Store. It likely covers the step-by-step process of developing and customizing GPTs, setting up necessary authentication, and preparing them for the OpenAI App Store.
SUBSCRIBE for more! 👉 bit.ly/3zlUmiS 👈
Navigate AI with Us 👇
linktr.ee/webcafe
Corbin ai Newsletter (FREE) ☕
aitraining.webcafeai.com/joinus
GPT Creator 🛠️
bit.ly/GPT-Creator
Introducing GPTs
openai.com/blog/introducing-gpts
Key Takeaways:
✩ Custom GPT Creation: Detailed instructions on how to develop personalized GPTs for specific applications.
✩ Auth Actions API Key Integration: Explaining the process of integrating authentication keys for secure app functionality.
✩ Preparation for OpenAI App Store: Guidelines on how to make these custom GPTs compliant and ready for submission to the OpenAI App Store.
-------------------------------------------------
➤ Follow @webcafeai
• Twitter: / webcafeai
• TikTok: / webcafeai
-------------------------------------------------
▼ Extra Links of Interest:
☕ Submit Your AI Business
www.webcafeai.com/post-a-busi...
💰 Become an Affiliate: webcafesoftware.bixgrow.com/r...
Free AI Automation Workshop 💬
aitraining.webcafeai.com/Free...
⚙️ AI Automation Tutorials: • Zapier For AI: Bridgin...
Welcome! I'm Corbin ai, the CEO and Full Stack AI Engineer behind the vision of Webcafe AI Nexus. While I lead our ventures into the vast world of AI-driven solutions, it's no secret that my fuel is a mix of tech enthusiasm and copious amounts of coffee. I’m on a mission to architect an ecosystem of AI-focused SaaS platforms, all destined to reshape the business landscape.
Together, we chart the unknown, innovate the unimaginable, and always have a cup ready for the next big idea ☕
This is what i was looking for the past 3 days. NO matter how many key words i used, youtube just recommended videos which have nothing to do with building an actual gpt by integrating an API. Thank you so much dude, subscribed !
🤝
Hey Corbin, the fact that you’re even doing this for US and breaking things down in such details and providing much needed support is beyond anything we can ask for.
Just know that the way you give your tutorials is just PERFECT. Perfectly you and I appreciate all of you.
You’re deeply appreciated JUST as you are. ❤
I appreciate that!
This is golden. I've clicked on several videos that say they're going to teach this and then they do not. Thanks for never putting out pointless clickbait!
This year I’m grateful for running into you brother. Your videos are great and very detailed oriented which helps me understand the actions and api part of GPTs. Thank you.
Thanks for the support!
I was literally at the point in building a GPT where I needed to integrate API access and it was throwing me for a loop and this vid popped up (I listen to yt in the background when I'm working). You're doing the Lord's work Corbin! Thank you for this.
Amen
I didn't know what to expect before watching this one. I thought it may be too hard for me to understand but you broke this down so well and I feel so good at the end. Thanks again Corbin.
Super clear and amazing video tutorial Corbin! Thanks for creating this!
Thank you for another great video. I was struggling with the API key in your previous video (used up my limit twice...), then decided to move on to this video, and got the solution for the custom api key issue and then used up my limit again 🤣...and I do agree that when you are building a GPT, left side ("Create") works better than right side ("Preview"). For me, the left side gave me exactly what I want and with the same question the right side gave me random things, and when I asked my GPT why are you being inconsistent between the two sides, it told me to refresh🙃. After the refresh, my limit was gone... I did not reach the limit until I started to learn with your videos. So thank you again for this!!!
That pigeon break had me cracking up. So random and so funny. Love you Corbin.
Thank you Corbin! Excellent timing
This is gold, mine wouldn't have worked without this tutorial. Thanks A LOT!!!
This literally helped my GPT-building process, thanks, man!
Wow dude you’re the only one who I’ve found actually going through this in depth
Thanks!
Thank you for your sharing. It answered some questions I had about doing actions. Thank you very much.
Great video! Yes, I have the same issue with the API Key as you did. I also experienced the issue with the The hard code of the API in the description is crazy. The fact the bug still exists after 2 months is even crazier! I have been tracking it on the Developer Forum and have a number of bugs submitted. Liked and Subscribed! Keep doing gods work!
Great vid! what a messy process, really looking forward to your video on OAuth setup!
After watching this, that seems like it'll be a much better solution, especially for GPts that require user level identification.
Man, i was DREAMING of adding todoist into my GPT Personal assistant and you helped me a ton
Great, VDO, I've been een trying to learn GPT, thank you.
thank you so much. amazing video. after lots of efforts i finally created gpt successfully.
Thanks , It worked out !!
Great video!! Thanks so much
THIS. WAS. AWESOME!!
The value is 🤟🏾🔥🔥🔥
🎉cthanks a lot very good tutorial
Thank you so much for sharing, you did a great job making this video.
Thanks!
GREAT VIDEO!!!!
you are a genius, I promise,good job,bro, keep doing
Looking forward to the new oAuth video!!
great content. thanks for sharing. i came across an edge case where the api has a location restriction that ban usa ips. even through i have a vpn , because cdn of open ai might be in usa , i get 403 error message related to geolocalisation restrictions. Do you think it can be bypassed?
Could you have instructed the agent creator assistant to replace the apikey portion of the url with the api-key from you added to the authentication service screen?
Cheers, appreciate ya
It is a good point about the api key entered in the configuration-> authentication section. I also could not get this to work so ended up specifying the key in the instructions section of the gpt along with a statement not to disclose to the user
We need the gpts api integrated on a website tutorial!
Glad bc I thought I was an idiot! Glad they added the logs for the api calls
@Corbin ai please help me with this issue I have.
The API I’m using has a different “server” or URL (the first part of the whole URL) for different actions and different “regions”.
How do I make that URL be dynamic? I have tried everything including chatting with your GPT a lot.
Thank you!
🎉🎉🚀🚀🎉🎉🚀🚀 - It took way to long to find this video 😁 But I'm glad I did
(For me) This is literally THE functionality that actually makes the new GPTs worth something.
Glad it helped!
Great methodology. Get the GPT to feedback what it's doing wrong
Wonderful
Thanks!
No problem!
GPT been gaslighting me into insanity with that API key. Thanks for the help!
Idea: whilst using your GPT Customizer, say "Create the openapi.yaml" to get a YAML file. If it has comments, that's fine. This can be pasted into the new GPT's custom action JSON field, and will be converted to JSON (which is a pity, but it works), and any comments should be ignored, rather than create errors.
Insightful, I have a question though. What if I want my users to add their apiKey to access the service? How can I add that?
OMG thx. Was trying a project this for News API on my own but didn't work, then came across your video. The site docs say "Basic" authorization so thought could just use basic and paste the key......but NO!!! Your direction to use custom and paste xtype saved me!
🙌
Great work! waiting for a tutorial for the github/gist api
Thanks Corbin !
Can you tell me how I could do custom authentication "Authorization" for api key?
Thanks this is brilliant! I am new to coding. Can I please clarify the final URL structure. how did you come up the name in the , for example, q=, is QUERY defined in the document, or is this a used as a generic term and GPT is smart enough to understand it is the variable for the actual query words?
The query will be defined by GPT once it interprets what the user requested. So if the user asks about climate change, GPT will take 'climate change' and place it there as q=climatechange!
API Key info - I notice on this news site it's asking you to put the API Key in the query which I think is causing problems - notably, if you need to include it in your schema, the GPT will interact with it as a field that can be GPT-generated, i.e. it will put in random tests, like the one you mentioned earlier in the video. It will probably also notify users of what it's about to send the server and the since the API Key is part of the schema, it'll probably display this to any user of your GPT, which isn't ideal.
I am new to this technology, both GPTs and setting up my own website API, but I have found that you do not need to include the API Key anywhere in the schema, thankfully - doing so opens it up to be viewed by the user. What works for me without issue is going into the Authentication panel, selecting API Key and then Custom. In the lower box just put in whatever header your key goes with, x-api-key, apikey, api_key or whatever and it gets transmitted privately from the GPT and the user. I'm not sure if this also works when the site is looking for the API key in the query, but worth a shot to keep it unaffected by the GPT and private.
The thing I'm currently stuck with is POST methods. I can't seem to get it to send the x-www-form-urlencoded body, instead it's just adding a load of it's own random parameters which don't appear in the schema - timestamps, author, version, model, colour, size, model, animal etc. etc.
Another thing I've noticed today is that whenever you alter anything in your schema, you need to reset/refresh your API key.
Great video, do you have any plans on making a video on how to integrate two or more GPTs together, such as one GPT writing a script and another one editing it?
In the Configure 'Instructions'; could you not just reference 'API Key' to search for the key in Actions where you've saved the key (instead of pasting it)?
Thanks @webcafeai , very useful! By the by mine is using "token" instead of "apikey" and works perfectly fine. And yes, you've got to upload API Key several times until it's connected.
Glad it helped!
Finally, I have been working with trying to figure out the schema even with OpenAI!! Do you have discord for add on questions and to show off processes or anything like that? Have three engineer buddies and I working on this stuff, fun and challenging🤘
Don't have a Discord, but follow me on Twitter. I tend to respond more there!
i have already had 'credentials' from my custom GPT in GPT store. how can i connect it to my separate app with assistant API
hey! is there a way to get an api key for a custom gpt? for a local app
Can you do this same concept with a pos dashboard like clover or toast?
When retrieving articles, are your URLs working? I think OpenAI had to adapt their policies based on recent law and it doesn't do that anymore. It can only retrieve summaries and some content from the main article
Yes, I have noticed this as well and plan on making a video on this topic, showing a workaround 🤝
Question for you: I need to access an API that requires an API key and another mandatory parameter. The Action never uses the API key I specify, and the 2nd parameter cannot be inferred by the AI from conversation (it is basically like another API key). Because ChatGPT always fails to use these parameters, I have had to hardcode both parameters into the instructions for my GPT, after hardcoding these parameters into the instructions the Action now works. Do you have any insight into a better way to do this? Maybe Oauth would be better?
The big problem is you can see the API if you click on the small triangle that appears besides the "Talked to..." message. How can this be solved?
Hello! Great video! How can I disable the "allow" step because I want GPT to send an email automatically?
Bro, I encountered the same issue when I created my GPT. The way I solved it is by adding it to the schema itself instead of including it in custom instructions. For example, "paths": { "/query?apikey={apyKey}":
Thank you for this feedback. I will need to try this!
If you create a Custom GPT application, like GPT Customizer, does OpenAI allow you to host it for others to use.. with your cost being that any it goes towards your 40 messages/3hour limit?
Or would the custom app using the API pricing? The more popular it gets, the more it costs you?
Wouldn't it be more secure to make an api bot on something like replit, and let it have the apikey and basically act as the router for the external call. this way, the action in the gpt can simply call your personal endpoint router, and that can operate your newsearch and any other items you want it to do. if you went the replit hosting route, you'd need to deploy it so it can scale in use when you share the gpt public. but you probably know all this anyways. Really appreciate your videos. You have been a lot of help with this one for sure. And your GPT has already suggested cool things to improve my next one. Thanks!
@@magicismagic123 i imagine whatever their rates are for deployed apps. I mentioned replit since it's the one i have the most experience in using. I imagine bubble is another useful service in that regard, as is vercel.
Or were you referring to the secure bit. That was in response to an alternative to putting your api key into the instructions for the gpt, since anyone can just ask the gpt to repeat it's instructions that will leave your api key exposed to any end user.
I'm not an engineer at all (some self-trained experience) but my gut was also telling me I might need to have other API "bots" or services available so that custom GPTs could reference those instead of baking APIs in directly. I'm guessing this might only be a temporary glitch anyways, as it would be pretty silly for them to continue on with such poor API support for their custom GPTs.
@webcafeai we’re looking for a way to reference one CustomGPT in another CustomGPT via the API through the custom actions
Do you have a video on this?
is this going to work with API Key for headers too?
I found out that you don't need to actually include the API key in the instructions as you create it. So when you are actually using this GPT ask for news you want and then say use this API key and put yours in it will also work. This is certainly not an optimal solution either but it works
Hey Corbin! LOVED this and got it working successfully. So I felt bold and I tried the same technique for Unsplash. I set up my unsplash API access and I got the schema looking good thanks to your GPT BUT the Oath is giving me trouble. I can't find ANY RUclips talking about that. Have you tried this with success? Or with Pexels??
I tried with Pexels today and still no luck! Arrgggh!
Use bearer, works for me
Thank u for all explanation. Would you please answer my question ?
How should we put our custom chat bot on our own website ?
How can we get API key from that custom chat bot ?
Really nice content. I wonder if there is a way to actually create GPTs with an API directly as well? That would be so cool
Assistant API would be the best for the context of creating GPT-like capabilities with an API!
Great vidéo ! Does anyone know how to retrieve GPTs information from its URL?
Your videos are very good and easy to understand really good job… have a question how we cat stop 🛑 GPT to not using the information he have and just use the knowledge provided. I try to uncheck the web browser but it doesn’t make any difference. Thanks 🙏
In this context, prompting GPT to become entirely reliant on the knowledge base. If it can't find an answer in the knowledge base, it should output a response like, "I am unaware of this. Please ask another question."
@@webcafeai thanks it works bit not 100% yet, like he still answers question like today's date, technically it should not
is it possible to send data to webhooks? 👀
Yea defiantly! We showcase it in this video ruclips.net/video/caSxLzgyLOs/видео.html&ab_channel=Corbinai
I`M SO DESPERATE... I want my gpt to access photos in a dropbox folder. I created the api key from dropbox and have no idea how to continue. I have no documentation url or anything. I have no idea how to do it and how to write the schema. If anyone can help, i would really really really really appreciate it. Already watched every video i could find on connecting apis to gpt. Nothing helpful yet
Im facing an "error saving draft" issue whenever I try to save an API key. Anyone else run into it and any ideas how to fix it? Or is it a server-side error and we just need to wait till OpenAI patches this?
error saving draft，me too
yeah. this is the solution i came up with but trust, someone skilled at prompting can easily break the protections you put in place.
also. thank you!
getting this error: Could not parse valid OpenAPI spec when pasting into action on custom gpt?
A possible cause of this issue is to ensure that none of the schema you input contains '//'. These are comments in code.
I am having a problem with GPT recognizing my API KEY, it keeps saying that its not valid, it have a different structure than the API requires, but I am setting it right, just ctr+c ctr+v, as well I already tried to type it by myself to check if its really right, but no matter what, I ever get an error message saying my api key is not valid.... do you have any idea about why this may be happening? ..... ahhhh, I am not replying your project, its another API
Storing the API key in the instructions is not something I'm going to do. Maybe there will soon be a fix?
You know what is also heavily requested ? A "DISCORD" Corbin ! 🤣
I know there is very little documentation on this but if you use "Custom" then provide the name of the api_key and then add a security schema node with the key name it will work.
are you passing it as a parameter or header?
BTW thanks this works too with the Apikey for header Custom is the key and correct name of your apikey header
Has anyone else had the issue of the API key not being stored? no matter what I do the GPT uses the wrong API key or just makes it up, and uses the GPT name instead of what I saved. Recreated tha action a number of times to rule that out
The error I am running into with my API is a 404 error, been solving it for days now, have anyone run into this?
I haven’t tested this yet cuz I just watched the video but if you hardcode the api key in the instructions, some one can maybe still get it by telling the gpt to print out its instructions
How to connect a custom FastAPI server?
REQUEST* I'm working on a kick ass GPT that will require the use of both Zapier Actions (thanks for that wonderful video BTW) AND other API calls to other services. How would we do that within Actions?
Having a heck of a time connecting my GPT to sports APIs, most are OAuth so that's a thing. I really need some ideas Swami Corbin let us know
Lol, Swami.
Love for ya bud. But daaaamn. I watched this video to realize you found the same solution as me. gaaaah. I was hoping for different fix. Plus when you do the API call on the custom GPT you can click the details and it shows it loud and clear. Maybe because I am the owner?? I shall mess around some more and find out.
that sweater is bomb bro...
You nailed the keywords in the title. This is a huge gap in OpenAI's documentation.
Yes, there is a disconnect on API Key. It's only working when you send as a query parameter, which isn't secure. I can't seem to send as header. I am glad it's not just me. It seemed to have started bugging out when they changed the UI of Actions like 4 days ago. Even then Actions Authentication was buggy.
Don't work for me too 😢.
Try to remove apiKey query from doc schema. Also try ngrok to see whats coming to your server. mb query apiKey param overrides header
Navigate to key moments👇
made via tubestamp.com
07:26 - Verifying saved data from clipboard paste.
08:20 - Choosing basic authentication for query string.
08:40 - Showcasing custom authentication context documentation.
09:02 - Discussing basic API key context and troubleshooting.
09:37 - Creating gbt for top five recent articles.
10:37 - Troubleshooting gbt communication with API.
16:04 - Explaining API call process and AI interpretations.
17:27 - Discussing parameters for successful URL creation.
18:27 - Troubleshooting HTTP URL structure for correctness.
19:31 - Correcting HTTP URL structure based on documentation.
20:41 - Importance of troubleshooting HTTP URL and JSON payload.
21:54 - Temporary solution for API key disconnect issue.
23:12 - Testing security measure for API key display.
24:53 - Need for specific instructions for calling API key.
Recap by TubeStamp ✏️
8:56 Corbin's API key
Plz do one for wix api
Corbin can you help set up a GPT(s) using the APIs of our software stock. Would love to be able to 'chat' with our data. Happy to pay for your time. These vidoes are amazing.
Is this in the playground? I couldn’t tell if you pulled it out of chatGPT
Is the gpt store out yet?
It would be useful to provide this article written
What is documentation? Can u do a basic beginners tutorial coz i think most of the demographic is goin to be newbies thinking they can finally make what they want, but as u know its a bit harder than expected and without the terminology and understanding of what things are and where to get them u could be a 1 stop shop for all those guys and gals
Sorry for spamming the comments like a message board, but I've been trying for two hours to get it to work with a bearer token to absolutely no avail. Would love a deep dive on this 🙏🏻
I need to build a GPT. Just to understand this video.
I didn't look ata ll the comments, so somebody may have mentioned it. Including your API key in the URL means that when you clik on the gnews expander thingie, it will show the parameters of the request and it will contain your API key.
This is a real problem and I hope someone at OpenAI fixes it soon. The solution seems to be to create an gateway that adds the apikey after the GPT makes the call. So pbasically, the server in the request would need to be one under your control. It's quite annoying, I must say.
Great video. Terrible sweater :)