The hesitation around AI on product data is usually about governance rather than capability. This demonstration covers how a Pimble is configured, tested and chained: choosing from ChatGPT, Claude, Gemini or a custom model trained on the business’s own data, setting where the output is written, building the prompt from the record itself, and testing in a sandbox before going near live products. Pimbles follow the approval workflows already defined, and chaining is conditional, so the right instruction runs against the right part of the catalogue.
With Pimberly’s Pimbles, you can:
- Choose from ChatGPT, Claude, Gemini, or a custom model trained on your own product data
- Build the prompt from the product record itself — attributes and media pulled in automatically, rather than typed each time
- Set a result attribute so the output is written to the right place on the product record
- Test a Pimble in a sandbox and restore earlier versions, before it ever touches live products
- Route generated content through your existing approval workflows, so a human reviews it before it’s accepted
- Chain Pimbles conditionally — so an image generation step can feed a marketing description step, and the right instruction runs only against the right part of the catalogue
See how it works in the video below:
