Hey HackerNews, for the past few years I've been exploring machine learning, from modeling all sorts of things, to language and vision models, all the way up to the other "consumer" end of the spectrum: using and crafting agentic apps. The learning curve has been steep, and the field moves fast. It's a lot for anyone to absorb.

I thought, having gone through this, can I use what I learned to make it easier for the person that comes next? That's where I am today.

With that in mind, I've started with open sourcing a project aimed at *simplifying and standardizing the usage of models, tools and agents*, so anyone can start coding and deploying cross-platform AI apps on day 1, without any prior AI experience, without learning engines or frameworks, and *on any hardware* (model, quantization, engine, backend all dynamically set by default, optimized for your device). The interface is later customizable, so it grows with you as you learn, up to production readiness.

This is all you need to get started:

=====================================================

from universal_intelligence import Model

model = Model()

result, logs = model.process("Hello, how are you?")

=====================================================

- Both local and cloud-based models are available, with the same simple interface. Can be instantly swapped.

- Similar APIs are made available for tools and agents —all of which are composable, interoperable, and capable of sharing hardware resources to scale at virtually no cost.

- Available in Python or JS/Typescript, usable anywhere (desktop, mobile, web, cloud).

- Can work with your favorite agent framework and MCP tools if you'd like.

Also, no more managing tokens or fetching “['choices'][0]['message']['content']”.

I'd love to hear about your experience and challenges, to think about where to take this next.