API - OpenRouter support
Hello, I’d fully be into this if there was API/OpenRouter usage so I could UP the models or even downgrade to whatever was desired if cost was an issue that week. It would massively increase flexibility of the tool and relieve you from having so many api requests yourself. I see someone else asked about AI model switching previously. 15 sites isn’t enough when a client wants 3 choices, that is 5 clients a month.
https://www.youtube.com/watch?v=bUt1WpDlI6E
Otherwise, I’m more inclined to just do it myself and mod a process like the one in this vid from Jay (or many other ways that came out in the last month with other tools to do it). If supporting more models is a pain in the neck, at least add Opus 4.7 and Sonnet and maybe one other on the cheap end like Deepseek or GLM 5.1. Many of us need options to support dozens of new clients per month x 3 choices usually (my usual SOP, they pick one, we mod it a bit to their liking then, and deploy). With basically a 15 limit, you are killing your sales, as this is basically a home use or small biz offer, and kind of NICHE at this point. There is no way an agency can use it. IF you’re limited to 5 clients or less (can’t count on 3 coming out correct each time either) you are limiting almost all your customers to 5 new clients a month tops. With GHL and leadgen stuff I can do that in a day, which leaves the biz with nothing to do for the other 19 or so days of work each month. IE you can fill your calendar with inbound/outbound email and calls 24/7 and wake up in the morning with 5-8 closers to make. A 1 person business with good leadgen can book a calendar for a month solid and need to make their own tools just to keep up with what the bots book all night while sleeping, and I’m talking a single person here let alone a team of 5-10 doing design. Numbers get ridiculous at that point.
Hope you have plans of expanding to allow model selection, as these days I don’t build a tool without at least openrouter and probably Anthropic on top (or openai stuff since they seem to be about equal these days).
Thanks for your time,
Aaron.
Log in to comment and vote
Comments1
Eliot Dill
May 14
Aaron, thanks for the well thought out feedback.
Just to clarify one thing, 15 sites means 15 completed sites, the credit estimates includes up to 3 theme choices for each of those 15 completed sites which actually equals 45 themes total, or 15 clients per month (not 5) to use your example.
That said, I do feel a sliding scale for additional credits at the agency level makes sense for larger or more volume-driven agencies so I’ll add that to the request list.
In regards to your actual feature request: OpenRouter
We’ve had BYOK requests but this is the first for OpenRouter. It is a good solution with some
caveats.
First, if we were to implement, it would only work for actual theme and page generation, not for other features that use credits as well. For example the site copy feature uses multiple services that utilize credits as does our forthcoming Preview on WordPress - both of these services use SaaS products that utilize AI but not with credits that are available via OpenRouter.
The next challenge we face in implementing this is the wide variance of output by different LLMs with the same input. We’ve tested this a bit against Gemini, Claude, and ChatGPT and the results differ quite a bit, which makes it even more challenging to then convert the output to a native WordPress theme in any predictable way.
Another challenge we have with this is credit transparency. Right now we use fixed credit counts for specific actions. The pros of this is that you always know how many credits you’ll use before you do. The down side is that credits aren’t 1:1 based on context window like OpenRouter and other LLMs charge (we have thresholds and limiters to control it). We may have to shift to a “context-window” and unpredictable credit usage model in order to offer this.
We’ve also evaluated adding a couple of direct options for LLM on the high (Claude/ChatGPT) and low end (Kimi/Deepseek), but that creates the same variance in point #2 above. Plus models like Opus and even Sonnet are significantly more expensive than something like Gemini. Instead of getting 15-45 themes you might get 3-7 total until we get more paying customers and can ask for a volume discount. That didn’t feel like it would serve our users well. They’d be happy to see Opus initially but then be disappointed when their credits disappeared very quickly.
Hopefully that makes sense. I really appreciate the feedback and suggestions. We will continue to look into this as it seems like something users want. There is just significant challenges to consider when actually implementing.