For serious applications, I think we need the ability to switch models within a Pickaxe.
For example, for system architecture, we might want the best Thinking from Sonnet, but for detailed implementation, maybe Gemini is a better value. Of course, the ideal would be that Pickaxe do automatic model selection like germerlin.in does pretty well (appears to be doing what I described in this example). If these guys from India can do it with no funding, I imagine, and challenge Pickaxe to do it with $3.8 million.
@nathanielmhld
@kenlyle have you tried connecting your Pickaxes in the Actions tab? If you set up multiple Pickaxes using different models, you can probably achieve a similar effect to what you’re describing above!
1 Like
Can you break that down a little more, or is there a doc?
I guess that Custom Action would be an option, but that would mean BYOK in the code.
So you are suggesting the Action that is like “Chain to Another Pickaxe”? In that case, the second Pickaxe is equivalent to any other action, as it has a Trigger Prompt?
Is there some architectural limitation that prevents us from being able to declare-
Image Preview Generation - ChatGPT Image 1.5
Image Editing - Nano Banana
Text Generation - Haiku
Something like that?
This would be a step toward a future auto-optimization process, possibly using LLM Leaderboard 2025 or similar, and a task specific internal value calculation of some kind by Pickaxe, cost per token/Leaderboard score, where lower is better.
I think that this is architecturally a bit more Agentic on the part of the platform, a bit like Gumloop which allows building agents vs. Make.com, which is more literal, step-by-step and brutish.
Such an architecture would be good for usability and would be a kindness to the early adopters who are Action-constrained.
In the bigger picture, is there any Roadmap? Is there any process by which we can actually get answers to reasonable questions like this, aside from Office Hours? The ideal answer would be of the form:
”Good idea. It’s been added to the roadmap for Q2 as Task ####. There are a few preliminary things that need to happen in Q1.” or even “This has been considered and rejected, because {reason}.”
Thank you for your thoughtful questions and suggestions.
To clarify, you can navigate to the Actions tab in the Builder and, within the Actions Library section, use the “Your Pickaxes” filter.
This will display all Pickaxes you have created. From there, you can add any of these Pickaxes as actions to your current tool. For each connected Pickaxe, you can set a specific trigger prompt to determine exactly when that Pickaxe (and its associated model) should be engaged during a conversation.
This approach allows you not only to use different models for different tasks, but also to have highly specialized Pickaxes work together within a single workflow. Each connected Pickaxe can be designed to handle a specific responsibility.
Furthermore, regarding image models, it has always been possible to connect different image models for separate tasks. For example, you can use one model for image editing and another for image preview generation. By providing clear and detailed instructions in your prompts, you can specify exactly when each image model should be used.
Thanks @danny_support
That is a very clear explanation, and we will try it out.
