Every other offline AI tool requires 6GB+ RAM — locking out billions of people. NanoMind runs on 1GB. Built for the other 3 billion.
Every existing local AI app — PocketPal, Google AI Edge, Off Grid — has the same requirement: 6–8GB RAM.
That locks out:
- ❌ Budget Android phones ($50–80 range)
- ❌ Raspberry Pi with 2GB RAM
- ❌ Old laptops with 4GB RAM
- ❌ Billions of people who can't afford flagship devices
NanoMind is built specifically for these devices.
| PocketPal | Google AI Edge | Off Grid | NanoMind | |
|---|---|---|---|---|
| Min RAM | 6GB | 6GB | 6GB | ✅ 1GB |
| Works on budget phones | ✗ | ✗ | ✗ | ✅ Yes |
| Works on Raspberry Pi | ✗ | ✗ | ✗ | ✅ Yes |
| 100% Offline | ✅ | ✅ | ✅ | ✅ Yes |
| Open Source | ✅ | ✅ | ✅ | ✅ MIT |
| OpenAI-Compatible API | ✗ | ✗ | ✗ | ✅ Yes |
| Platform | Status |
|---|---|
| 🐧 Linux / Raspberry Pi | ✅ Available now |
| 🍎 macOS | ✅ Available now |
| 🪟 Windows | ✅ Available now |
| 📱 Android APK | 🔨 Coming in v0.2 |
| 🍏 iOS | 📋 Planned |
Step 1 — Clone and install
git clone https://github.com/Minhajul-Mahib/nanomind
cd nanomind
pip install -r server/requirements.txtStep 2 — Download a model (auto-picks best for your RAM)
python scripts/download_model.pyStep 3 — Start NanoMind
python server/server.py
# → Running at http://localhost:8080
# → API docs at http://localhost:8080/docsStep 4 — Chat with it
from openai import OpenAI
# Works with ANY app built for OpenAI — just change the base_url
client = OpenAI(base_url="http://localhost:8080/v1", api_key="not-needed")
response = client.chat.completions.create(
model="nanomind",
messages=[{"role": "user", "content": "Am I running locally?"}]
)
print(response.choices[0].message.content)
# → "Yes. 100% on your device. Zero data sent anywhere."| RAM Available | Model | Size | Quality |
|---|---|---|---|
| 1GB+ | TinyLlama 1.1B | 0.6 GB | Basic |
| 2GB+ | Gemma 2B | 1.5 GB | Good |
| 3GB+ | Phi-3 Mini ⭐ | 2.2 GB | Great — recommended |
| 6GB+ | Mistral 7B | 4.1 GB | Excellent |
The installer detects your RAM automatically and picks the right model. No manual hunting on HuggingFace required.
Your Device (1GB RAM minimum)
┌─────────────────────────────────────┐
│ NanoMind Server (Python) │
│ ┌─────────────────────────────┐ │
│ │ FastAPI │ │
│ │ OpenAI-compatible API │ │
│ │ /v1/chat/completions │ │
│ └──────────────┬──────────────┘ │
│ ┌──────────────┴──────────────┐ │
│ │ llama-cpp-python │ │
│ │ 4-bit quantized inference │ │
│ │ CPU-only — no GPU needed │ │
│ └──────────────┬──────────────┘ │
│ ┌──────────────┴──────────────┐ │
│ │ GGUF Model (local file) │ │
│ │ TinyLlama / Gemma / Phi-3 │ │
│ └─────────────────────────────┘ │
│ │
│ ⚠ Zero internet after setup │
│ ⚠ Zero telemetry ever │
└─────────────────────────────────────┘
- v0.1 — Python API server (OpenAI-compatible)
- v0.1 — Auto model downloader (detects RAM, picks best model)
- v0.1 — Linux / Mac / Windows / Raspberry Pi support
- v0.2 — Android APK ← in progress
- v0.2 — In-app model browser
- v0.3 — Voice input/output
- v0.3 — Chat with local files (RAG)
- v0.4 — iOS support
- v1.0 — Plugin system
We especially need:
- Testers — run it on your device and report RAM usage + tokens/sec
- Android developers — help build the APK
- Translators — make NanoMind accessible in more languages
git clone https://github.com/Minhajul-Mahib/nanomind
cd nanomind
pip install -r server/requirements.txt
python server/server.pySee CONTRIBUTING.md · Join Discord
Most AI progress benefits people with expensive hardware. NanoMind is built on the belief that private, local AI should work for everyone — not just people with flagship phones.
If you have a $50 Android phone, you deserve AI that respects your privacy too.
Built by @Minhajul-Mahib · nanomind.ink · MIT License
⭐ Star this repo if you believe AI should work on every device