How do you use the Assistants API?

Curious how do you use Assistants API?

Which tools do you use the most, is the most useful?

Which tools do you use the least?

3 Likes

I’ve been playing and discovering a lot with the Functions tool. I feel that its capabilities are still in the early days and will be really impressive in the near future.

About the RAG, although it works really well, its still to expansive to keep using without making profit of it

1 Like

I developed a framework on top of assistant API. What I can do right now is amazing, I feel like I did time travel. I’m software engineer and at the moment with that software I built, I do not code anymore, I just discuss with the AI by voice, no more keyboard and mouse. It scaffold my projects entirely, refactor very complex codebase, debug complex problem and solve. At the moment for real world use case, as I use it on my everyday work, he beats my 20 years of experience in the field in some aspect. We work as a team and I streamline part of my job to it while I’m at something else. I will probably release a video and article to share my discoveries.

7 Likes

I personally use Assistants API literally, meaning as my personal assistant. While none of the agent frameworks on the market seem to be production ready (Assistants API not excluded), they are extremely useful as productivity assistants.

One of my use cases is using this API for prompt engineering - once I’ve done a test of say 100 cycles, I then feed the results to the assistant to check them (instead of going through them manually).

1 Like

This sounds amazing. I’m waiting for more info!

I have used Assistants api and built an open-source UI for creating an assistant and adding to anyone’s website or embed it anywhere GitHub - SamurAIGPT/Open-Custom-GPT: Create Custom GPT and add/embed on your site using Assistants api

2 Likes

We use it here internally for a lot of qualification jobs (evaluate incoming pitches against our investment criteria, check the company website to get more information , check pitchbook. Then send an email if missing needed information is missing and then repeat.

We also use it to create weekly reports based on a few salesforce queries. I love that one most because literally just takes the json with the results for last week and the results from the week before and creates a write up about how the week went. This is all the code it takes using the wrapper I wrote.

4 Likes

getWeekReport is the function that gets called by OpenAI as part of the Assistant. weekReport submits the task in the Celery queue and when its done the sendWeekReport function gets called to send the email.
As you can see the getWeekreport function is really just taking the results of a few Salesforce queries and throws in a Dict. (Note that some of the queries are groupings in order to make sure the total size of the response is never crazy. What I really love is how I can just provide two ‘records’ in this Json, one with last week and one with week -2 and the rest is up for openAI. What is even better is that I do NOT have to define a set structure BEFOREHAND in the Function definition. OpenAI doesn’t really seem to care what kind of JSON it gets back, it just ingests it.
email example

I wrote about and share the code (not as a library yet) without paywall here:

2 Likes

I would love to see a video of you in action. It is quite amazing what can be done but have not seen it being used the way you discussed.

1 Like

Hi. Do you have a live example on a site somewhere? Love to take a look at it. Thanks.

1 Like

There is a cloud hosted version for the project here to try https://customgpt.thesamur.ai/ and test it out if you want to check it quickly without self-hosting the project locally

Please, can someone help me. I need an OpenAI forum, where they speak Spanish. This message is translated. I don’t know English. Thank you

Did you host a node js backend webservice api to integrate with openai assitant api?

Check your website in browser dev tools, if in network fetch/xhr you can find the openai assistant api request, and find your apikey in request header Authorization…, it would be a disaster.

Be cautious, openai’s apikey should be safely hidden at backend, or you will pay for anyone who use your apikey.

I am not familiar with next.js. Maybe I am wrong, but check it twice, it’s not a small issue. openai javascript sdk is used in node.js, not in front end js framework…

The api key is not entered by admin, it is entered by the user using the custom gpt and is local to the user and not shared with other users using the custom gpt. It’s completely secure as your api key never leaves your browser

So, target user is he who has an OpenAI apikey and deposit some credit in OpenAI in order to use the assistant api and paid to OpenAI for himself?

I thought Assistant API is for commercial integration to provide service to common, public consumers, and for those anvanced consumers who has ChatGPT Plus subscription, we can use GPT builder on OpenAI’s GPT platform directly to provide our AI stacks, which don’t need api fee and is more performant.

Those who seek for Assistant API usually want to deliver AI for public audience, if they misunderstood, then they expose their apikey to public. So you may need to caution them not to misuse?

Assistant API is not to be integrated in any production environment at least for now, while it is still in beta.

Hi @craig.indylimotransp
I am preparing to release the Brainstack (it is a set of frameworks) with my AI companion iBrain. I’ve been working on it for the past 6 years. It’s a cognitive architecture in which ChatGPT is a part of as the complex reasoning component. However, it’s having more components like awareness, mind, memory and much more. It’s complementary. In the last few weeks, I’m working on a non linear cognitive process allowing creating, learning. At first I was using Assistant API, but now I’m in process of developing something more simple yet doing more. Kind of a task manager with a “universal workflow”. I wrote dozen of papers, POCs, codes all privates. I’m about to disclose everything soon. Here is a screenshot of the paper I’m completing. drop you email, I’ll share a copy :slight_smile:

How long til you release it, do you think? I just watched your videos and am very intrigued to try it! Just give me a place to sign up and I will be the first to give it a try :slight_smile:

Hey @daniil

You can subscribe at iBrain One. However, I want to mention the website is under active development and is unreleased for now, so don’t judge on what you see at the moment. Use it only for the subscription. :slight_smile:

Mart