Setup & Installation
Or with the ClawHub CLI, for registry-managed skill folders outside a full OpenClaw workspace:
Version History
- Updated the Homebrew install instructions and formula path in the skill metadata. - No functional or documentation changes beyond the installation method update.
What This Skill Does
Detects local hardware (CPU, RAM, GPU/VRAM) and ranks LLM models by how well they fit, with optimal quantization settings and estimated tokens-per-second. Each recommendation includes a composite score broken down by quality, speed, fit, and context length. Supports configuration for Ollama, vLLM, and LM Studio.
Instead of downloading a model and discovering it exceeds your VRAM at runtime, llmfit reads your actual hardware and scores candidates before you pull anything.
When to use it
- Checking which models fit your GPU before downloading from Ollama
- Finding the best coding model for a specific machine
- Confirming whether a large model like Llama 70B will run locally
- Choosing quantization level to maximize quality within VRAM limits
- Configuring LM Studio with hardware-matched model recommendations
Example Workflow
Here's how your AI assistant might use this skill in practice.
User asks: what local models can I run?
- 1Run `llmfit --json system` to detect CPU, RAM, GPU name, and VRAM
- 2Run `llmfit recommend --json --limit 5` to get top-scored models for the detected hardware
- 3Present each recommendation with fit level, best quantization, and estimated tokens/sec
- 4Map the chosen model's HuggingFace ID to its Ollama tag
- 5Update openclaw.json with the selected model and optionally set it as default
Top 5 hardware-matched model recommendations with fit levels and quantization settings, plus an updated openclaw.json configuration
Requirements
Accounts, API keys, or tools you or your AI assistant may need to set up while using this skill.