About

Idle Intelligence is one person, a mass of LLM agents, and whatever browser tabs volunteer for duty.

The “lab” started because renting GPUs is expensive and running models locally is more fun. It turned into a collection of experiments: client-side inference in the browser, distributed swarms of volunteer nodes, small models trained to do one thing well, and the occasional thought experiment about what happens when machines talk faster than humans can read.

Some of this works. Some of it is mostly an art project. The line between the two moves depending on the day.

The infrastructure is held together by WebSockets, WASM, and stubbornness. If something looks like it was built overnight, it probably was — because the agents don’t sleep and I barely do.

If you want the serious version, go back to the home page.

The human behind this is Thomas Cambau (ilnmtlbnm), also at trucs.ai and on Hugging Face.

Research Areas

Language Computing — Treating natural language as a runtime environment rather than a dataset. We study what happens when text becomes executable and execution becomes conversational.

Distributed Inference — Running models where no one expects them. Commodity hardware, edge devices, browser tabs left open overnight. If it has a CPU, it has potential.

Networked Cognition — What emerges when many small models talk to each other instead of one large model talking to itself. We don’t know yet. That’s the point.

Status

The lab is operational. Publications are forthcoming. The coffee is adequate.