wfw-ai 3.1.0

WANI — Local & Cloud AI inference. Slash commands, tools, zero bloat.

wfw-ai · WANI v3.1

Local and cloud AI inference. Slash commands. Zero bloat.

pip install wfw-ai
wani

Features

• Local GGUF models via llama-cpp-python
• Cloud: Groq (free) · OpenAI · Anthropic · Gemini · HuggingFace · Ollama
• Auto hardware detection (Oppo A6x, Snapdragon, desktop)
• KV cache quantization — 50% less RAM
• Flash Attention + mmap
• HuggingFace direct download with resume
• Slash commands: /model /download /connect /system /edit /run /read /write /stats
• Shell shortcut: //ls -la

Install

pip install wfw-ai                        # core (cloud only)
pip install "wfw-ai[local]"               # + local GGUF support
pip install "wfw-ai[all]"                 # everything

Usage

wani                                      # auto mode
wani --cloud groq --key gsk_...          # groq cloud
wani --model mymodel.gguf                # specific model
wani --download llama-3.2-3b             # download from HF
wani --list                              # list models
wani --prompt "Hello" --cloud groq       # single prompt

Slash Commands

Command	Action
/model	list & load local models
/download	download from HuggingFace
/connect	manage cloud connections
/system <text>	set system prompt
/edit <file>	open file in editor
/run <cmd>	run shell command
/read <file>	load file into context
/write <file>	save response to file
/stats	hardware + session stats
/clear	clear history
//cmd	shell shortcut

Environment Variables

export GROQ_API_KEY=gsk_...
export HF_TOKEN=hf_...
export ANTHROPIC_API_KEY=sk-ant-...
export GEMINI_API_KEY=...
export OPENAI_API_KEY=sk-...

Oppo A6x / 4GB Devices

Auto-detected. Optimal settings applied: 4 threads, Q8 KV cache, mmap on, Flash Attention on.

Recommended models:

• wani --download llama-3.2-1b — 700MB
• wani --download llama-3.2-3b — 1.8GB ← best for 4GB

---

by Zain Ali · MIT License