signerless/llm-checker

What it solves

Choosing the right local LLM is difficult because users must manually balance model size, quantization, and their specific hardware limits (VRAM, RAM, memory bandwidth). LLM Checker automates this by analyzing a user's system and recommending the best-fitting models from a massive registry of over 33,000 artifacts.

How it works

The tool uses a hardware detection engine to identify GPUs (NVIDIA, AMD, Intel, Apple Silicon) and CPU/RAM capabilities. It then applies a deterministic scoring system across four dimensions—Quality, Speed, Fit, and Context—to rank models. It integrates with Ollama for model management and execution, and includes a WebAssembly-based structural safety validator (ModelVet) to ensure model files are safe to load before they are executed.

Who it’s for

It is designed for developers and AI enthusiasts running local LLMs who want to optimize their hardware utilization without manually calculating memory requirements or searching through thousands of model variants.

Highlights

• Multi-Source Registry: Accesses 33k+ artifacts from Hugging Face, Ollama, and GPT4All.
• Hardware-Aware Recommendations: Calibrates suggestions based on actual VRAM and system memory.
• MCP Server Integration: Allows AI assistants (like Claude or Cursor) to analyze your hardware and manage local models via the Model Context Protocol.
• Model Safety: Includes built-in structural verification for GGUF and safetensors files to prevent loading corrupted or malicious files.
• Performance Metrics: Provides live tokens-per-second tracking during model execution via ai-run.