Releases·Tool Release
Dev launches LexiPanel, a self-hosted control plane that fits llama.cpp, image, audio and ONNX AI workloads to your hardware and auto-optimizes them
A developer released LexiPanel, a self-hosted control plane that fits AI models to the user's actual hardware and workload rather than just launching them. It manages llama.cpp, stable-diffusion.cpp, audio and ONNX runtimes from one browser, benchmarks real workloads, auto-optimizes, and requantizes models to VRAM budgets. It targets homelab and small-fleet operators, positioning itself underneath apps and agents rather than replacing Ollama, vLLM, or Open WebUI.
13 engagement1 sourceFri, Sep 25, 2026, 05:15 PM
AI Overview
- LexiPanel runs multiple independent llama.cpp, stable-diffusion.cpp, audio.cpp, Camelid and ONNX Runtime instances from one browser, with CPU/GPU/NPU support including AMD Ryzen AI, Intel and Qualcomm NPUs
- Offers 220+ explained controls, shows launch commands, VRAM/RAM estimates and refusal reasons before start, and prevents unsafe launches by reading GGUF metadata and resident workloads
- Auto-fit benchmarks quant formats on actual GPUs, builds tensor-level requant plans to a real VRAM budget, validates changes against real traffic and rolls back regressions
- Measures long-context decode behavior, thermals, power and tokens/joule; supports AMD GPU auto-revert of unstable tuning settings
- Ships an OpenAI-compatible gateway with users, API keys, quotas and usage accounting, plus fleet mode where multiple boxes share models through one gateway with failover
- Exposes 28 MCP tools, supports Hermes Agent config generation and llama.cpp Web UI, and runs on Python stdlib only with files plus systemd
- Builder positions it under apps and agents, not replacing Open WebUI, Ollama, GPUStack, LocalAI or vLLM
Related
Products·building
llm-tune: tuning tool for running local LLMs on your hardware
llm-tune is an open-source tool that helps people find the best settings (quantization, GPU offloading, cache, sampling) for running local LLMs on their specific hardware. It benchmarks configurations across GPUs and engines like llama.cpp, Ollama, and vLLM. It solves the trial-and-error problem of getting good local model performance for local LLM hobbyists.
llm-tune is an open-source tool that helps people find the best settings (quantization, GPU offloading, cache, sampling) for running local LLMs on their specific hardware. It benchmarks configurations across GPUs and engines like llama.cpp, Ollama, and vLLM. It solves the trial-and-error problem of getting good local model performance for local LLM hobbyists.
4 engagement1 sourceredditFri, Oct 9, 2026, 02:49 AM
Social·Community
Engineer shares experience running LLMs on Nimo Strix Halo system
A hardware engineer reports using a Nimo Strix Halo system for local LLM inference, noting it cannot run the largest open-weight models or high-quantized 100-200B models but is sufficient for internal tools, Python scripts, and CI/CD pipelines. The post highlights the practical trade-offs of using consumer-grade hardware for AI workloads.
A hardware engineer reports using a Nimo Strix Halo system for local LLM inference, noting it cannot run the largest open-weight models or high-quantized 100-200B models but is sufficient for internal tools, Python scripts, and CI/CD pipelines. The post highlights the practical trade-offs of using consumer-grade hardware for AI workloads.
16 engagement1 sourceredditFri, Jul 17, 2026, 10:21 PM
Releases·Tool Release
Quartermaster: open source local AI platform, a llama-swap fork, auto-configures GGUF models behind one OpenAI-compatible API
A developer announced Quartermaster, an open source local AI platform that began as a fork of llama-swap and has since grown into its own project. Users point it at a models folder and it auto-configures runtime settings by reading GGUF headers and measuring free VRAM. It targets ease of use while keeping all settings editable per model.
A developer announced Quartermaster, an open source local AI platform that began as a fork of llama-swap and has since grown into its own project. Users point it at a models folder and it auto-configures runtime settings by reading GGUF headers and measuring free VRAM. It targets ease of use while keeping all settings editable per model.
44 engagement1 sourceredditThu, Sep 3, 2026, 07:36 PM
Products·launched
Local Model Explorer: checks which GGUF quants fit your GPU or Mac, with llama.cpp commands
A Hugging Face Space that ranks ~3,000 popular GGUF models by whether they fit a user's hardware, given context length and KV cache settings. It reads each GGUF's header to compute memory use instead of guessing from parameter counts, and outputs a runnable llama.cpp command. It helps people running local LLMs on GPUs, Macs, Strix Halo, or DGX Spark avoid downloading quants that won't fit.
A Hugging Face Space that ranks ~3,000 popular GGUF models by whether they fit a user's hardware, given context length and KV cache settings. It reads each GGUF's header to compute memory use instead of guessing from parameter counts, and outputs a runnable llama.cpp command. It helps people running local LLMs on GPUs, Macs, Strix Halo, or DGX Spark avoid downloading quants that won't fit.
7 engagement1 sourceredditThu, Sep 17, 2026, 03:13 AM
Social·Community
User seeks sanity check on headless 7800XT desktop serving LLMs to Mac for vibe coding
An IT professional is using a headless desktop with 7800X3D CPU and 7800XT GPU to serve LLMs via llama.cpp to an M1 Pro MacBook for coding. They run Qwen3.5 9B and Gemma 4 12B locally on the Mac, and seek community feedback on their setup.
An IT professional is using a headless desktop with 7800X3D CPU and 7800XT GPU to serve LLMs via llama.cpp to an M1 Pro MacBook for coding. They run Qwen3.5 9B and Gemma 4 12B locally on the Mac, and seek community feedback on their setup.
11 engagement1 sourceredditSat, Jul 25, 2026, 06:32 PM