← use cases · Edge & embedded teams
Edge inference appliances
Ship an inference server as a single Multiboot ELF that boots on the metal — no distribution to patch, no init system to harden.
The problem
An inference appliance built on a general-purpose Linux image carries a package manager, an init system, a userspace, and a supply chain of dependencies you did not write. Each is attack surface and each adds boot latency you do not want on a device in the field.
Why cllm fits
- ▸The deployment unit is one Multiboot ELF — nothing else on the box to update or exploit.
- ▸Boots in milliseconds from a Multiboot loader, on QEMU or bare-metal x86.
- ▸A 4 MB heap arena and a handful of files keep the resident footprint tiny.
- ▸Every line between the NIC and the model is code you can read and audit.
Honest caveats
- ▸The llama.cpp inference engine is not integrated yet — the appliance serves the API skeleton today.
- ▸Only i386 is targeted; ARM64 edge boards are future work.
- ▸No persistent storage: models are baked into the binary or loaded at boot.