Projects — mudler.pm

NAME

projects — things mudler has built, ported, maintained, or broken over ~two decades

CREATED

LocalAI Open-source AI engine — LLMs, vision, voice, image and video on any hardware, no GPU required. (Go · MIT · 47k stars)

LocalAGI Self-hostable AI agent platform; drop-in replacement for the OpenAI Responses API. (Go · 1.8k stars)

LocalRecall 100% local memory layer and knowledge base for agents, with a web UI. (Go · 900+ stars)

EdgeVPN Decentralized peer-to-peer VPN built on libp2p — no central server, token-based discovery. (Go · 1.9k stars)

Kairos Immutable Linux meta-distribution for edge Kubernetes. Donated to the CNCF. (Go · CNCF · 1.7k stars)

luet Zero-dependency container-based package manager using SAT solving and QLearning. (Go · 300 stars)

yip YAML instructions processor — a lightweight cloud-init alternative. (Go)

poco Portable containers — turn container images into static, daemonless binaries. (Go)

golauncher Extensible application launcher and window switcher in under 300 lines of Go. (Go)

docker-companion Squash and unpack Docker images, in Go. (Go)

entities Declarative modern identity manager for UNIX systems. (Go)

MCPs Personal MCP servers used with LocalAI. (Go)

LOCALAI ECOSYSTEM

apex-quant Adaptive precision for expert models — MoE-aware mixed-precision quantization. (localai-org · 400 stars)

privacy-filter.cpp OpenAI's privacy-filter NER model in a minimal C++/ggml runtime. (C++ · localai-org)

LocalVQE Neural real-time acoustic echo cancellation — ggml and PyTorch inference. (C++ · localai-org)

vibevoice.cpp C++ port of Microsoft VibeVoice on ggml. (C++ · localai-org)

ced.cpp ggml port of the CED audio-tagging model. (C++ · localai-org)

moss-transcribe.cpp C++17 port of OpenMOSS MOSS-Transcribe-Diarize. (C++ · localai-org)

free-splatter.cpp ggml port of FreeSplatter pose-free Gaussian splatting. (C++ · localai-org)

rf-detr.cpp RT-DETR object detection in C++ with ggml. (C++ · localai-org)

voice-detect.cpp C++17 speaker recognition and voice analysis engine. (C++ · localai-org)

GGML / C++ PORTS

depth-anything.cpp From-scratch C++17/ggml port of ByteDance's Depth Anything 3. No Python at inference. (C++ · 800+ stars)

parakeet.cpp NVIDIA Parakeet ASR in C++ with ggml. (C++ · 700+ stars)

locate-anything.cpp NVIDIA LocateAnything-3B ported to ggml. (C++ · 350+ stars)

voxtral-tts.c Pure C implementation of Voxtral-4B-TTS. (C)

vllm.cpp Community 1:1 vLLM port in C++. (C++ · active)

LIBRARIES & TOOLS

cogito Go library for co-operative agentic software, tuned for small OSS models. (Go)

nib Zero-dependency terminal LLM agent harness — a single static binary. (Go)

skillserver Create, manage and share skills between agents. (Go)

go-llama.cpp The original Go bindings for llama.cpp. (go-skynet · 900+ stars)

go-pluggable Light bus-event driven plugin library for Golang. (Go)

herd An embeddable runnable DAG for Go. (Go)

Mojo::IOLoop::ReadWriteProcess Perl process management library, donated to openSUSE. (Perl)

Algorithm::SAT & Algorithm::QLearning SAT solving and Q-learning on CPAN. (Perl)

CONTRIBUTED TO

Gentoo Linux Gentoo developer — genkernel-next, MATE maintainer.

Sabayon Linux Lead developer of the distribution and its community repositories.

openSUSE / openQA Rewrote the openQA scheduler and its process-management layer at SUSE.

Cloud Foundry KubeCF and Quarks-operator work at SUSE; EiriniX extensions.

llama.cpp Created the Go bindings; assorted fixes.

gpt4all Contributed the Golang bindings.

MODELS

LocalAI function-call fine-tunes Function-calling fine-tune family — llama3.2 1b/3b, phi-4, qwen2.5-7b, llama3-8b. (Hugging Face)

Italian LLMs Minerva-3B instruct tunes (Minerva-3B-Llama3-Instruct, Asinello-Minerva). (Hugging Face)

TALKS & PAPERS

Secure Edge-Native Architecture (with Intel) Whitepaper on trusted boot and secure edge with Kairos.

Livin' Kubernetes on the (Immutable) Edge The New Stack, on Kairos.

Trusted Boot at the Edge The New Stack, on securing devices at the edge.

LocalAI meets k8sgpt CNCF talk.

Building immutable Linux derivatives with cOS Conference talk on the Elemental toolkit lineage.