What if large language models transform from cloud APIs, and become something we coordinate across the machines we already own?
Distributed Intelligence
DevDen is a research journal focused on software craftsmanship, distributed agents, open weight inference, local runtimes, and the future of everyday compute.
We believe the next chapter of LLM infrastructure will not be written only inside massive GPU data centers. It will also emerge from the idle capacity already sitting around us: laptops, desktops, phones, home servers, gaming machines, and edge devices.
Long Tail Inference
We study architectures where model execution can be split, routed, or scheduled across devices rather than always centralized behind a cloud endpoint.
Projects like Petals, exo, vLLM, and llama.cpp point toward different parts of this future: peer inference, local clusters, high throughput serving, and efficient local runtimes.
Edge LLMs
We look at systems where language models run closer to the user, improving privacy, latency, ownership, and resilience.
As open weight models are changing the economics of language model deployment. DevDen explores how these models can be compressed, quantized, served, composed, evaluated, and verified in real world systems.
The question is not only which model is best, but where it should run, how its outputs should be trusted, and how much of the stack developers should own.
The Future Compute Fabric
We imagine a living network of everyday devices that collectively forms a new layer of global compute.
