ISOCAST is a personal AI lab running entirely on hardware at home — a mesh of nodes serving local language models, image & video generation, and autonomous agents.
No cloud bills, no rented GPUs. Just metal, fans, and a lot of late nights tuning it.
Compute pooled across the lab — used for model inference, training experiments, and rendering.
Each box has a job. Together they run an LLM mesh, generation pipelines, and a wall of background workers.
Everything self-hosted — the goal is to own the whole stack, end to end.
Dozens of open models served across the nodes — code, reasoning, vision and embeddings — load-balanced for parallel work without a single cloud token.
Diffusion pipelines (Krea / Qwen-Image on ComfyUI) generating high-res images locally on the GPUs.
Text-to-video (LTXV) producing short clips on consumer hardware — up to 1080p, right at home.
Swarms of coordinated agents handling research, code, and review — running around the clock in the background.
Custom monitors keeping an eye on every node, GPU and worker — so nothing runs silently into the ground.
The whole estate is firewalled, patched and monitored — a lab that's locked down as carefully as it's built.