OpenAI Assistants API Tutorial - Playground & NodeJS

[@samsquamsh78]

Fantastic Leon! Really like how you don't rush and explain each step and avoid taking shortcuts! Well done!

[@leonvanzyl]

Appreciate your feedback 🤗

[@m4rtin419]

It’s a few weeks in since Assistants were released and this is still the video that explains it best!!! Would be great to great to see more Assistants API videos from you!

[@leonvanzyl]

Thank you very much. I'll definitely create more videos on Functions, Code Interpreter and Retrieval.

[@Kutsushita_yukino]

damm its only been that long?

[@nickmills8476]

That was a great intro to the Assistants API. Concise, and easy to understand.

[@ClownCallout]

You are the best! I became a Flowise god using your tutorials, now Flowise has added OpenAi assistants in version 1.4.0

[@leonvanzyl]

Thank you very much, you made my day with this comment 🤗.

[@nicolasgramnea1324]

​@@leonvanzyl you have one more here that you helped a ton 👋

[@musumo1908]

@leonvanzyl adding to the love here!

[@The_Outkeeper]

One more

[@ClownCallout]

A pleasure! ... You made my decade! So grateful for your brilliant straight to the point clear tutorials I am now on the path to greatness! I am joining your discord I have a Mulit $Million Idea I want to share with you.

[@sebastiandeanda2085]

This was a great demonstration of how to run the Assistants outside the playground. I haven't seen any other videos though explaining how to hook up the actions to the assistant.

[@leonvanzyl]

Thank you for the feedback! Exactly, no one is going to use this in playground only. We WANT to integrate this into our applications.

[@lyda1283]

Thanks for such a useful video. It's taken me days to find a video that actually shows you step by step how to create and import the assistant. And you explained it in such an easy to understand way.

[@leonvanzyl]

You're welcome 🤗

[@JoeD0403]

Great video! People may make some interesting assistants directly within OpenAI, but what you explained is the key to making amazing and practical assistants.

[@leonvanzyl]

Thank you for the feedback!

[@kirilkirchev285]

Thanks for the comprehensive approach in your videos. They always have straight to the point approach. I am a great fan of your channel since I discovered it in April this year. I have a question - do you use some tools when creating an audio for a video?

[@leonvanzyl]

Amazing comment! Thank you! 🙏. I don't really use any SFX in my own videos, so can't really give a recommendation 😅

[@brilliant332]

Thanks Leon, I'd love to see how you work with tools and functions and how to bring some decision making into the AI.

[@leonvanzyl]

Will do 👍

[@bigbadcatbigbcy2933]

This tutorial is so clear. You can get everything you need to do with openai api. Thanks a lot

[@leonvanzyl]

Thank you 🙏

[@matheus.maiberg]

You are an awesome person, the world needs more people like you. I appreciate every bit of effort that you put into this video. Thank you so much!

[@leonvanzyl]

Thank you for the kind words 🙏

[@AleXzorZ]

Yes please, more videos on the assistant API! This guide is the first one I've seen that actually got me coding along (I'm noob) 8-) I feel like i gradually started to understand the logic of the language. Thanks!

[@leonvanzyl]

That is awesome! Glad you enjoyed it. I'll definitely continue this series. We still need to look at programmatically adding files to the knowledge base, and custom functions.

[@Kira-fj8el]

This video is the best tutorial that I found for open ai assistant but there is small issue that encounted. That is when i try to run whole code onece, the reply msg is not retrive and remaining code is being executed. When next time fun again, previously retrived answer appered, due to time delay of the api responce

[@leonvanzyl]

Thank you for the compliment!

[@Kira-fj8el]

@@leonvanzyl can u resolve that issue from a video

[@leonvanzyl]

@@Kira-fj8el yes, I'll continue with the series looking at practical examples. If I understand your issue correctly, you are unsure of how to get the AI response after running the agent. In a nutshell, you have to fetch the latest status of the run in a loop until a completed status is received. After this you can fetch the list of messages to retrieve the AI response.

[@codearabawy]

Thank you very much! I watched the DevDay keynote and your video to understand the practical side of it. Would really be helpful if you make a video for calling assistance apis functions too.

[@leonvanzyl]

Will do

[@NickSealPueo]

Thanks this was helpful. I couldn’t figure out why I kept getting assistant api errors until watching this.

[@leonvanzyl]

You're welcome 🤗

[@riotov1706]

Hey I’m trying to use the retrieve method to retrieve an assistant I made on the open Ai website but I’m getting an error saying the openai.beta.(assistants).retrieve Assistants in brackets because that’s what it’s saying does not exist

[@leonvanzyl]

I don't think the SDK changed (my own implementations of the solution are still working). Why is assistants in parentheses in your example? Be sure to check the docs as well. platform.openai.com/docs/assistants/overview?context=with-streaming

[@joshbrowndev]

Thanks Leon! Great video and fantastic explanations. Would love to see more Assistants related content

[@leonvanzyl]

Noted. I appreciate your comment. Helps to know what I should focus on. Anything I should cover?

[@joshbrowndev]

@@leonvanzyl Anything regarding best practices working with the API in real world use cases. Integrating into a UI, writing instructions, using tools, current limitations (eg streaming). I’m still very new to this so any content that provides a better understanding of working with the API in a practical sense would be very valuable, at least for me. Thanks again!

[@NewUser12345-i]

Hello, with this openai update, what do you consider to be the advantages of using flowise with pinecone vs building everything with open ai assistants api?

[@leonvanzyl]

The Assistants API does not offer an easy solution for embedding it in apps and websites. You could use Flowise with the Assistants node for this. Also note that the Assistants API has a massive limit on the Knowledge Base. Think you can only load 10 or so files, where Flowise (Langchain) does not have these limits. The Assistants API is just not ideal for larger enterprise solutions in its current state. This could change of course, but until then there is still a need for frameworks like Langchain and AutoGen. Vector Stores like Pinecone does not have a limit on the amount of sources that you upload.

[@NewUser12345-i]

@@leonvanzyl Thanks! One other important question: if I build an application on top of flowise. Do I have any risk in case flowise goes out of business or if it's hacked? I think it's not possible to export code

[@codewithtae6637]

This is good stuff. Appreciate the video! Keep them coming!

[@leonvanzyl]

Will do

[@personone6881]

My "node app" doesn't appear in my list of Assistants when I refresh the browser window. I have been meticulous following along however, the only thing I can see that is different between your screen and mine (apart from the project directory, naturally) is that in the Terminal, my path starts with: """ (base) PS C:\etc... """ Where-as yours simply reads: """ PS C:\etc... """ What have I screwed up? :/

[@personone6881]

Nevermind... Solved: I had it the file in my OneDrive project directory... (I moved it to a directory on my ACTUAL "C:\etc." - at least I think that was the issue - doing so seemed to fix it :))

[@leonvanzyl]

Glad you got it sorted 👍

[@samnelson9702]

cant get it working for me, for some reason I cannot get it to create a new assistant or fetch an existing one, no communication between the two. I tried different API keys, made sure everything was exact to your video and still cant get it working.

[@leonvanzyl]

What's the error?

[@andrescondo62]

Currently, there is a bug, which takes a while to respond, since the run remains in an "in_process" state. Do you already know how to solve it?

[@alex_turing]

So after you do run the run can have a status of queued you said. Does this mean retrieving the new message at that moment is not possible? Do we have to somehow wait for status to be complete before getting the new message? If so what way do you suggest to do it? Use an interval to check every second if the status is complete or something? And the assistant can send multiple new messages. So how do we know what the new messages are to only send those to the user? Do we iterate the list till we meet his message?

[@leonvanzyl]

That's correct. You cannot send new messages while run is incomplete. You might notice on OpenAI Playground, that they actually disable the ability to send messages while the run is busy You might want to implement something similar in your app.

[@alex_turing]

@@leonvanzyl I've created a system where each user has his own thread. However, I've noticed I get this error often: InternalServerError: 500 The server had an error while processing your request. Sorry about that! at APIError.generate (/home/container/node_modules/openai/error.js:65:20) at OpenAI.makeStatusError (/home/container/node_modules/openai/core.js:263:33) at OpenAI.makeRequest (/home/container/node_modules/openai/core.js:306:30) at process.processTicksAndRejections (node:internal/process/task_queues:95:5) at async useChatGPT (/home/container/chatGPT.js:46:19) { status: 500, headers: { 'alt-svc': 'h3=":443"; ma=86400', 'cf-cache-status': 'DYNAMIC', 'cf-ray': '854ff8299fae4bc9-BUF', connection: 'keep-alive', 'content-length': '175', 'content-type': 'application/json', date: 'Tue, 13 Feb 2024 20:54:56 GMT', 'openai-organization': 'user-bdcywrsyy5ryvdhirztdzweb', 'openai-processing-ms': '65', 'openai-version': '2020-10-01', server: 'cloudflare', 'set-cookie': '__cf_bm=MRPGCHelicgR.JMSiDoHRvBTESgbCYG5n7_PyzC8OSk-1707857696-1-ATcgUKolTD9tplyk4tkTTA7E9E3xyTf1GaSFUWE+ygjiygiUcsUSLAMLcbmGiiCJZjIb3+bCA/GvMurlOtSxRqM=; path=/; expires=Tue, 13-Feb-24 21:24:56 GMT; domain=.api.openai.com; HttpOnly; Secure; SameSite=None, _cfuvid=rBZbxj8loSlW2Zv6v7oAGzIkEeSlsELg6tSY508k7q0-1707857696379-0-604800000; path=/; domain=.api.openai.com; HttpOnly; Secure; SameSite=None', 'strict-transport-security': 'max-age=15724800; includeSubDomains', 'x-request-id': 'req_2c92df819af026c3b6cafbdfbe2ce0ef' }, error: { message: 'The server had an error while processing your request. Sorry about that!', type: 'server_error', param: null, code: null }, code: null, param: null, type: 'server_error' } It happens when I make some request to chat gpt, like to make a new message in a thread. I think it happens always when two users want to send a message at the same time, the first user will be successful while the second will get this error. As if OpenAI cannot service two threads in parallel. Some times though I get this error even when only one user requests a message

[@amajid0268]

Excellent video. Are you able to do a walkthrough of using custom api that has oauth secure authentication, used in an assistant. Using nodejs to execute the assistan

[@leonvanzyl]

Thank you 🙏. I'll look into this.

[@InLifeGPT]

Great Work Leon!!! Keep it going👍

[@leonvanzyl]

Thank you 🙏

[@faizaniqbal3629]

Hi Leon! I have one more confusion. If I understand correctly, as an AAA agency, our primary objective should be to master the development of Assistant APIs. Additionally, we require a cost-effective platform that allows us to embed these assistants into websites, with a view to offering a more affordable solution than existing chatbot builders like Voiceflow, Botpress, and Flowise AI. Please let me know if my understanding is incorrect

[@leonvanzyl]

You're understanding is correct. First, OpenAI Assistants include Retrieval and Memory built in. This is a significant cost reduction as you don't have to sell additional services likes vector stores to your clients. You can then select a mechanism to include the assistant in an app / website like Botpress, Voiceflow,.Flowise, etc. It's important for AAAs to understand this tech.

[@faizaniqbal3629]

@@leonvanzyl thank you for your answer.I really appreciate it.

[@traxtibidox5775]

I was waiting for this video from you 😂 ❤️ Thank you so much for this Leon 🙏

[@leonvanzyl]

Whoohoo 🤗. Glad I could help!

[@AliKibao]

I've been waiting for your take. Thank you.

[@leonvanzyl]

You're welcome 🤗

[@TAMIMMAHMUD]

Best video on the assistant API. Period 👏

[@leonvanzyl]

Thank you

[@asrulmunir]

Can we use Function tool to call external APIs?

[@leonvanzyl]

Absolutely! I'm working on a function calling video where we'll use the function to call an external API.

[@bendingreality-]

This is great! Is this possible to set up in Flowise too? (in the graphic interface)

[@leonvanzyl]

Yes, already created a video on that as well 😄

[@bendingreality-]

@@leonvanzyl ah lol, I'm blind. Awesome, just watching :)

Nice tutorial, Leon! Simple and effective! Thank you!

[@leonvanzyl]

You're welcome 🤗

[@tigana]

So incredibly helpful

[@leonvanzyl]

You're welcome 🤗

[@ConAim]

What is the limitation on this? Not that they will allow everyone with just basic package can write websites spamming their services????

[@leonvanzyl]

Not sure I understand the question, but there isn't a limit. You are charged for the tokens used, so if you don't have rate limiting and "spam" the service then you'll simply end up with a massive bill. OpenAI will be quite happy 😊

[@OndrejDusek-w4d]

Love this vid, Leon. Thank you! Learning only from you these days :). In case you're scouting for follow-up topics - I'd love to see how to use the new OpenAI assistant in Flowise in complex and multistage-agent setups :)

[@leonvanzyl]

Thank you for the feedback and suggestions 🤗. Glad I could help.

[@LucasLaino]

Great work!!

[@leonvanzyl]

Thank you

[@marcelo5688]

It´s amazing. Everything worked. Thank you

[@leonvanzyl]

You're welcome 🤗

[@danielgibbon2463]

Hey Leon. Loved the video. I just wanted to ask what the difference is between the OpenAI assistants API and something like flowise. Because to my knowledge, you can set a very similar system to an OpenAI assistant with vector stores in langchain. Does using the OpenAI assistance API actually provide any advantages or is it equivalent to just setting up a vector store in something like flowise?

[@leonvanzyl]

Hey there. I am going to release a new video in the next few days that combined Flowise with Assistants API. That should hopefully answer your questions. In short, Assistants offer long term memory and Vector Search out of the box. No need for paying for services like Pinecone, etc. Assistants API does not provide an easy solution for embedding the bot in your website. This is where combining Flowise with Assistants API is the perfect fit.

[@danielgibbon2463]

@@leonvanzyl Can't wait! I think Liam Ottley showed how to set something up like this with Voiceflow (kzbin.info/www/bejne/gZ-ZnGmlmKqUgZYsi=hleedIyIQ2jE9vlx) and I just wanted to ask how Flowise would compare to Voiceflow for this kind of application?

[@khalidkifayat]

@@danielgibbon2463considering cost effectiveness, flowise is opensource & free, while VF comes with 50$/month. FW UI is not that mature but can do tasks comparing VF.

[@insight-guy]

Thanks. Always appreciate your insights.

[@leonvanzyl]

You're welcome 🤗

[@oOserkanCakmakOo]

Thanks for sharing useful informations its so clear

[@leonvanzyl]

You're welcome

[@khalidkifayat]

Leon u simply rocked. just a hypothetical question. What should be the asking price of this maths tutor project, if a client needs it and wants me to make it for him ?

[@leonvanzyl]

Thank you 👍. You'll probably want to also embed the agent on the clients website, apps, etc. The cost should include the effort to set up the agent as well as the effort to and costs for these tools (Flowise, Botpress, Voiceflow, etc). Or maybe the client wants something completely custom with user authentication and subscription plans. It's really hard to give you an answer based on these points.

[@khalidkifayat]

@@leonvanzylif all your points are included, what should be a decent price ?

[@alex_turing]

@@khalidkifayat I would tell the client to use his own API key that he can pay to increase the credits it has. Also I would tell the client to use his own server service for hosting it. Then he should only pay for your efforts. And in my opinion this is something you can make in one day so the price could be pretty low. But if there is demand and not any other people providing it then you can ask for more than it's actually worth. I wouldn't ask for more than 50 euro to be honest

[@habertvler8363]

does it works if we create assistant with files uploaded in chatgpt ui, and then just run a new thread with that assistant id, would it work? ( i think so )

[@leonvanzyl]

Interesting question. The file needs to be associated with the specific assistant. Uploading the file via ChatGPT will not automatically assign the files to the assistants.

[@Kira-fj8el]

No it works for me

[@Kira-fj8el]

We can upload files via assistant gui

[@alexeyefimik5710]

Great tutorial, thank you! Please, make more! :) I wanted to point out how you can improve your videos: Please take a bit more time for typing of the code, i had to go back so many times.

[@alexeyefimik5710]

+ it was a bit hard to follow the code as it was constantly under the pop-out windows 12:09

[@leonvanzyl]

Thank you very much for the feedback. I'm always looking for ways to improve

[@faizaniqbal3629]

Hi Leon! can you please answer this? What is stopping OpenAI from introducing a subscription model in the future, where for a fee like $50 per month, businesses could effortlessly integrate GPT into their websites or WhatsApp using a GPT builder without any coding? What potential barriers could deter OpenAI from rolling out such a service? As AI garners increasing attention from the business community, will competitors likely follow suit with this monetization strategy? Moreover, what could be the consequences for AI automation agencies as this trend advances?

[@leonvanzyl]

As a AAA you should stay up to date with the service offerings by OpenAI and their competition. Honestly, nothing is stopping OpenAI from releasing those services. However, clients will always be willing to pay for convenience and your knowledge and experience.

[@faizaniqbal3629]

@leonvanzyl, this implies that in this business approach, there's always the possibility that OpenAI might turn into a competitor at any point. I agree with your view that numerous clients are ready to invest in integrating these systems efficiently into their businesses. I had thought earlier that Open Ai would not offer such a service as it needs a very large database to store the user information and knowledge or similar technical issues which is stopping them from doing so.

[@greyhairchooselife]

defaq, you enlighten me a lot!

[@leonvanzyl]

You're welcome 🤗

[@nicolassuarez2933]

Outstanding!

[@leonvanzyl]

Thank you

[@juanmartinkuasnosky55]

Hi leon, great video!!! do you have this repository on github??

[@leonvanzyl]

Thank you! I'm working on a comprehensive video. Will include all source code in that project.

[@DanielNicolasFrenchCookies]

I do not see Assistant in my OpenAi. I have GPT4

[@leonvanzyl]

I'm not aware of any restrictions. Ensure to use the link on the description.

[@mhb11]

Can an assistant spawn (and later destroy) another assistant?

[@leonvanzyl]

Guess it's possible. You could add functions to add and destroy agents. These functions then call the APIs.

[@matthewkk5087]

You exceeded your current quota, please check your plan and billing details. I am getting this error. should I buy something? on the other hand your video is an amazing

[@leonvanzyl]

Hey there. That means you've exceeded your OpenAI credit limit, or your free credits have expired.

[@TinyMagiq]

Good one @leon a create in playground and tie it with code.

[@leonvanzyl]

Thank you 🙏

[@mikew2883]

Excellent overview! 👍

[@leonvanzyl]

Excellent comment 👏. Thank you 😊

[@Ryan-yj4sd]

I’d be interested to see the Flowise integrations

[@leonvanzyl]

Working on it

[@XCmdr007]

Makes two of us. Would you Leon say that using Flowise to access Langchain functionality is still powerful? I saw your comment about using it as a Wrapper which makes sense if the use case is to use it for chatting. Langchain’s functions go beyond this as we mall know and Flowise provides for lots of cool stuff. What is your take? …. And … would you be working on a video that shows how to create actions with the Assistant API … and perhaps something from Flowise? 😅 Thanks so much for your awesome content. It is done with much clarity and wisdom. The impact you have on folks like me is transformative. Thank you again!

[@leonvanzyl]

@@XCmdr007 I use both Flowise and Langchain all the time. If you're not a developer, then Flowise is a fantastic low-code solution for creating and selling flows to your clients. For bespoke solutions I use Langchain (mostly the JS version). I'll be covering Assistants API with Flowise SOON, and will definitely create a video on Actions / Functions as well 😁.

[@nicolasgramnea1324]

​​@@leonvanzyl Flowise needs to be deployed on the cloud to embed a chatbot on a live website? Or how does they work? I looked many of your videos but this part is not clear to me, as I see you work mainly on local host. Thank you for your time.

[@leonvanzyl]

@@nicolasgramnea1324 I have a video on deploying Flowise that you should out then 👍. For these tutorials I simply run Flowise on Localhost, but for real world, production applications, you need to run Flowise in the cloud. For example, if you want to embed Flowise in a website (which is online) then you cannot use a local endpoint.

[@Kondrill]

Hi Leon, great content! Subscribed directly! Do you think it is possible to use this Assistants API Beta and create an UI with Chainlink with it? I've tried it, but I have to admit that I am not so long in the coding business.. Anyways, thank you for your Videos!

[@leonvanzyl]

Thanks for the compliment and sub 🤗. I'm not familiar with Chainlink, so don't want to risk giving you an answer. I might work on a video showing how we can create a UI for Assistants API.

[@Kondrill]

Ah that's great to know and thank you for your quick answer! @@leonvanzyl

[@anow2]

It's a shame that the Assistant builder doesn't have the same features at ChatGPT's "GPT Builder" - importing an API by just providing the OpenAPI YAML was magical.

[@leonvanzyl]

Agreed

[@musumo1908]

Awesome video! Thank god for people who do proper coding to explain everything 😉😄 Do you think this will get rid of needing memory stores and vector dbs for simple cases?? Im now able to run amazing stuff from just the openai assistant node in. FW….although not checked my bills yet

[@leonvanzyl]

Thank you for the feedback! This will definitely replace the need for seperate memory and vector stores. This is still in beta, so your results may vary, but it's exciting to know that the results will improve.

[@musumo1908]

Yeah for sure! I see you can actually upload a doc into threads from the openaidocs…if you’re ok with a closed system it’s now way easier to build stuff…will def kill a lot of those endless startups lol…now to find a cool frontend to wrap all this in!

[@alexismatzopoulos6930]

Thanks Leon!

[@leonvanzyl]

You're welcome 🤗

[@gerdberlin2]

More of that please!

[@leonvanzyl]

Will do. Working on retrieval and custom functions. Stay tuned 😁

[@Pure_Science_and_Technology]

CAUTION! FYI there is a failure of direct retrieval with GPT-4 using the new OpenAI Assistant API. GPT tokenizes text and creates its own vector embeddings based on its specific training data. The new terms and sequences may not connect well to the pretrained knowledge in GPT's weight tensors. There was no semantic similarity between the new API terms and GPT's existing vector space. This is a fundamental issue with retrieval augmentation systems like Rag - external knowledge is not truly integrated into the model's learned weights. Adding more vector stores cannot solve this core problem. The solution is to have multiple learned "knowledge planes" with trained weight tensors for specific tasks that can be switched in. This is better than just retrieving separate vector representations.

[@leonvanzyl]

Interesting. The solution is still in beta, so I'm sure this will be improved soon enough.

[@BitsNBytesAI]

Excellent

[@leonvanzyl]

Thank you

[@leonardosouzaconradodesant6213]

Super like

[@Hana-qf5sn]

Thanks

[@leonvanzyl]

You're welcome

[@Lucaslima-gz9vk]

you need to make 9237 requests to get a simple response :(