@avasis-ai/inspect

Hardware inspector for Avasis Agent Builder. Scans your device and recommends the best small models that run locally.

Install

npx @avasis-ai/inspect

What it does

• Detects CPU, RAM, GPU (VRAM), OS, and disk
• Matches your hardware against a catalog of 12 small models (0.6B to 14B)
• Recommends models that fit your device with GPU-accelerated or CPU-only ratings
• Outputs a Config ID you can use to pre-fill the agent builder at avasis.ai

Usage

# Interactive scan with colored output
npx @avasis-ai/inspect

# JSON output (for automation or builder pre-fill)
npx @avasis-ai/inspect --json

Example output

  ┌─────────────────────────────────────────────┐
  │        Avasis Device Inspector              │
  └─────────────────────────────────────────────┘

  DEVICE
  ─────────────────────────────────────────────
  CPU     Apple M1 Pro
  Cores   10 physical / 10 logical
  RAM     32.0GB total, 18.2GB free
  OS      macOS 15.0 (arm64)
  Disk    456.2GB free of 960.0GB

  RECOMMENDED MODELS
  ─────────────────────────────────────────────
  Model                 Size      Speed     Quality   Fit
  ──────────────────────────────────────────────────────────────
  qwen3:0.6b            0.4GB Q4  fast      basic     CPU
  phi4-mini:3.8b        2.4GB Q4  fast      good      CPU
  gemma3:4b             2.8GB Q4  fast      good      CPU
  ...

  BEST MODEL FOR YOUR DEVICE
  ─────────────────────────────────────────────
  phi4-mini:3.8b
  No GPU detected, runs on CPU

  NEXT STEPS
  ─────────────────────────────────────────────
  1. Install Ollama:     https://ollama.com/download
  2. Pull your model:    ollama pull phi4-mini:3.8b
  3. Build an agent:     https://avasis.ai/builder

  Config ID: a1b2c3d4 (use this to pre-fill your agent)

Supported models

| Model | Parameters | Quantized Size | Min RAM | Best For | |-------|-----------|----------------|---------|----------| | qwen3:0.6b | 0.6B | 0.4GB Q4 | 2GB | Classification, simple Q&A | | phi4-mini:3.8b | 3.8B | 2.4GB Q4 | 4GB | Code assist, file tasks | | gemma3:4b | 4B | 2.8GB Q4 | 6GB | Code generation, multi-tool agents | | qwen3:4b | 4B | 2.6GB Q4 | 6GB | Reasoning, structured output | | llama3.2:3b | 3.2B | 2.0GB Q4 | 4GB | Instruction following, tool calling | | mistral:7b | 7B | 4.4GB Q4 | 8GB | Complex reasoning, code generation | | qwen3:8b | 8B | 5.0GB Q4 | 8GB | Deep analysis, multi-agent supervisor | | phi4:14b | 14B | 8.5GB Q4 | 16GB | Expert reasoning, full autonomy | | qwen3:14b | 14B | 9.0GB Q4 | 16GB | Research, long-context tasks | | gemma3:12b | 12B | 7.5GB Q4 | 16GB | Multi-modal, supervisor agent | | llama3.1:8b | 8B | 4.9GB Q4 | 8GB | General purpose, agentic tasks | | deepseek-r1:7b | 7B | 4.4GB Q4 | 8GB | Reasoning, math, logic |

How it works

1. Uses systeminformation to scan hardware (cross-platform, no native deps)
2. Compares your specs against a model catalog with real size/requirement data
3. Rates each model as GPU-native, CPU-only, or incompatible
4. Picks the best model for your device

License

MIT