⚡ AMPLIFY

A local-first Small Language Model chat platform with a pluggable inference backend and a private analytics plane.

Runs 100% offline · Deployable in 2 commands · Portfolio-ready

---

✨ Why AMPLIFY

Most "local LLM" projects lock you into one inference stack. AMPLIFY treats the inference engine as a plug-in: switch from HuggingFace Transformers to LM Studio to on-device Termux/llama.cpp by changing one environment variable.

• 🧠 Pluggable backends — HF · LM Studio · Termux (llama.cpp)
• 🔒 Local-first — no data leaves your box
• 📊 Private analytics — Streamlit dashboard, password-gated
• 🐳 One-command deploy — docker compose up
• 🏠 Homelab-ready — designed to run on an 8 GB laptop / mini-PC

---

🏗️ Architecture

                    ┌─────────────────────────────────────┐
                    │            nginx :80                │
                    │   /  → app   /analytics → dashboard │
                    └───────┬────────────────────┬────────┘
                            │                    │
                 ┌──────────▼───────┐   ┌────────▼─────────┐
                 │  Flask app :5000 │   │ Streamlit :8501  │
                 │  ┌────────────┐  │   │  Plotly charts   │
                 │  │ Strategy:  │  │   └────────┬─────────┘
                 │  │  hf │ lm  │  │            │
                 │  │  studio   │  │            │
                 │  │  termux   │  │            │
                 │  └────────────┘  │            │
                 └───────┬──────────┘            │
                         │                       │
                         ▼                       ▼
                ┌────────────────────────────────────────┐
                │  ./data/amplify_chat_history.csv       │
                └────────────────────────────────────────┘

See docs/architecture.md for the sequence diagram.

---

🔌 Backend matrix

Backend	Env value	Best for	Deps
HuggingFace	hf	Fully local, offline homelab	requirements-hf.txt (torch)
LM Studio	lmstudio	Desktop GPU users	requests only
Termux	termux	On-device Android inference	requests only

---

🚀 Quickstart

cp .env.example .env
docker compose up -d --build

Then open:

• Chat UI → http://localhost/
• Analytics → http://localhost/analytics/ (basic auth)

Local dev (no Docker):

make install
make run          # chat on :5000
make dashboard    # analytics on :8501

---

🧪 Tests & Quality

make test    # pytest + coverage
make lint    # ruff + black --check
make format  # auto-fix

CI runs on every PR: Ruff, Black, Pytest with coverage, and a Docker build check.

---

📜 License

MIT © 2026 Araf Mustavi