I made the video below, demonstrating how to integrate AI with a Knack app, using Make and OpenAI.
It’s pretty simple to do, and can be done with a free Make and OpenAI account.
Once you’ve mastered this basic technique, a whole world of possibilty opens up for extending your app:
- Summarize text (as shown in video)
- Moderate content automatically
- Clean up messy data
- Generate images based on user input
- Answer questions from your users
- And so much more!
Let me know what other AI-related tutorials you’d like to see and I’ll consider making more videos!
Excellent video and use case
Thank you for sharing.
Hi Callum, excellent work. It will certainly be very useful for the Knack community.
A tutorial that involved a risk matrix using AI for prediction would be interesting.
Example: based on historical data from a certain number of variables (4 or 5 fields) which risk classification (low, medium, high) can be assigned to a certain action (doing something wrong, delaying a project, damaging a process , …) or object (person, company, profile, …)
We have started to use the AI after watching your video. We are using it to give our support requests a basic title. We do need to edit occasionally but it does save us quite a bit of time.
Thank you for the video. I would love to see other videos for more ideas to use AI with Knack.
Nice @CSWinnall ! I’m glad you found it helpful.
Some of the more in-depth AI integrations I’ve built have required a lot of experimenting with the prompt engineering - there’s so much to know about how to ask the AI for exactly what you want (which I didn’t go into here).
Some things you might try to get better results and have to manually adjust less often are:
- Giving some examples in your prompt of the sorts of titles you like and don’t like. Your prompt could end up quite long, which is OK.
- Changing the “temperature” parameter (an advanced action in the openAI Make step). The lower the temperature the less “creative” (less random) and the higher the temperature the more creative (random). When you need a highly predictable result, a lower temperature of, say 0.1 is much better.
The testing of different prompts, temperatures etc can take a bit of time - lots of trial and error, but can really be worth it.