Forex-Rates Plugin and Custom GPTs (Free and premium) for Chat-GPT from Forex-GPT.ai, available for free on the Chat-GPT Plugin Store to Chat-GPT Plus subscribers.
Background:
I remember writing about AI and making related predictions as far back as 2016, including discussing this on a panel in 2017, but it wasn’t until recently that I jumped on the AI bandwagon in March 2023, after upgrading to the premium version of OpenAI’s Chat GPT service.
API and early plugin access: In April I started using the API to run Auto-GPT, and then got notified of early developer access to Chat GPT Plugins in May, which includes the ability for you to develop and use your own plugin on the Chat GPT interface.
Coding a plugin over a weekend: I got access on Friday and started coding over the weekend and by Sunday night had a working version that pulled forex prices from a broker API, and performs some rudimentary technical and sentiment analysis on currency market prices.
Forex-Rates plugin for Chat-GPT: I named the plugin Forex-Rates and it’s available on the Forex-GPT.ai website, it was a cool experiment of what is possible and things to come. I used a technique called Meta-Prompting to help code more effectively. The software architecture is as follows:
It’s worth noting that due to the token limitation on Chat-GPT, you have to be mindful of the ratio of granularity to the total duration you ask for, as certain requests can inevitably be too large for Chat-GPT to handle.
For example, if you are asking for one-minute data on the EURUSD, anything more than a few hours of data can trigger an error due to the maximum message limit, when Chat-GPT retrieves the prices from the broker API.
This limit is because there is a lot of data contained in those API requests (i.e. open, high, low, and close prices, and descriptors), but you can ask just for bid prices, for example.
https://user-images.githubusercontent.com/5213035/240931027-4e029080-9b86-44fc-a3a1-b347045b1844.mov
Example ways to interact with the plugin:
“What is the overall sentiment for the USDJPY currency pair using bid prices on weekly candles for the past sixty weeks, excluding volume data?”
If you would like to learn more or contribute to the project, the code is open-source and can be found on GitHub.