Questions engineers ask first.

cllm is a bare-metal C unikernel for serving large language model inference over HTTP. It is Multiboot-compliant, boots in QEMU or on bare-metal x86, and exposes an HTTP/1.1 server with a llama.cpp-shaped REST API. There is no host operating system — the kernel and the application are one ELF binary.

A Multiboot ELF that boots, brings up serial and VGA, walks the PCI bus, drives an Intel e1000 NIC, and answers HTTP/1.1 requests through a llama.cpp-shaped v1 API. The inference engine itself is on the roadmap; the kernel and the network stack ship now.

Not end to end. The v1 endpoints exist, parse requests, and route to handlers, but inference returns a stub until the llama.cpp engine is integrated into the kernel. Treat cllm today as a working substrate that the inference path will land on.

The API surface mirrors llama.cpp, so the natural target is anything llama.cpp loads — GGUF-format weights for Llama-family architectures, Mistral, Qwen, Gemma, Phi, and so on. Until the llama.cpp inference path is wired in, no specific model has been benchmarked end to end.

QEMU is the development and primary supported target. The kernel is Multiboot-compliant, so any Multiboot loader will boot it — including GRUB on bare metal. Other hypervisors are unverified.

GPU support is roadmap, not shipped. The documentation includes an analysis of how a CUDA ggml backend could be integrated into a unikernel — PCIe access, PTX kernel embedding, host/device memory management — but that work is not yet in the build.

Raw Ethernet frames over the e1000 driver, with a minimal IPv4 + TCP implementation and an HTTP/1.1 subset on top — enough to route the v1 endpoints. There is no socket API in the traditional sense; the server is the packet-processing loop.

Only x86 (i386, 32-bit Multiboot) is wired up today. The kernel boots as a 32-bit Multiboot ELF in QEMU or on bare-metal x86. x86_64, ARM64, and RISC-V are noted as future considerations but are not implemented.

Install gcc with -m32 support, make, and qemu-system-i386. Clone the repo and run make run — serial output appears on your terminal in under a second. Ctrl-A X exits QEMU. There are debug, VGA, and GDB-attached variants as well.

No. The infrastructure scaffolding — kernel, drivers, HTTP — is in place; the inference engine is not. cllm is a research-grade substrate, not a production serving stack.

A unikernel removes the kernel/userspace boundary, the scheduler, and every page of the host OS that is not part of the inference path. The target is a single tens-of-kilobytes ELF that boots in milliseconds and spends every cycle on math.

cllm is open source under the GPL. The canonical source and LICENSE file live at github.com/cognisoc/cllm.

Cognisoc is the organization developing cllm. It is one of several open-source LLM inference projects under Cognisoc, each targeting a different language or device profile. The full portfolio and documentation live at cognisoc.com and docs.cognisoc.com.