Show HN: Clusterflock: An AI orchestrator for networked hardware
TL;DR
Clusterflock is an open-source AI orchestrator designed to manage agents across distributed hardware with varying VRAM and RAM constraints.
Key Points
- It automatically profiles networked hardware and downloads the best-fit models from HuggingFace without manual configuration.
- Native parallelism via llama.cpp enables multiple smaller models to run simultaneously on the same device.
- The built-in mission runner supports multi-session, asynchronous agent workflows out of the box.
Nauti's Take
This is exactly the kind of tooling the self-hosting community has been missing. Cloud providers solve the deployment problem by throwing money at it – Clusterflock solves it with smart hardware-aware logic instead.
Whether tight packing holds up under real-world load remains to be proven, but the architecture looks thoughtfully designed by people who actually ran into these problems themselves. Worth watching closely.
Context
Running AI agents on self-hosted hardware typically means manually distributing models, juggling VRAM limits, and coordinating deployments across machines. Clusterflock tackles this bottleneck with hardware-aware automation that was previously only practical in cloud environments. Being fully open-source and built on llama.
cpp makes it especially compelling for self-hosters and small teams who want serious agentic infrastructure without vendor lock-in.
Sources