✦ Aether — private, local AI for everyone

Like the invisible medium once thought to fill all of space — an intelligence that runs quietly, everywhere, on your own machine.

A clean, friendly chat interface for running AI entirely on your own computer. No cloud. No API keys. No sign-up. No data collection. Nothing you type, say, or show it ever leaves your device.

Works with any model you install through Ollama — Qwen, Llama, Mistral, Gemma, Phi, and more — and adds the things that make a local model actually pleasant to use: image understanding, voice, and a hands-free Live mode.

🎧 Live mode — just talk. It hears when you stop, thinks, and talks back — then listens again. Talk over it to interrupt. Like Gemini Live, but 100% private.

👁️ It can see. Attach any image and ask about it — the app switches to a vision model automatically:

---

💡 Why local?

	Cloud chatbots	This
Your data	Sent to a company's servers	Never leaves your machine
Cost	Monthly subscription / API fees	Free
Internet	Required	Works fully offline
Accounts	Sign-up, logins	None
Models	Whatever they offer	Any open model you choose

Perfect for private notes, sensitive work, flaky-internet situations, or just learning how modern AI works without a bill.

---

✨ What it can do

• 💬 Chat with any local model — swap models from a dropdown, no restart
• 👁️ See images — attach, paste, or drag in a photo and ask about it
• 🎧 Live voice mode — hands-free conversation with talk-back and barge-in
• 🔊 Read replies aloud with your computer's built-in voices
• 🎙️ Talk to it — your speech transcribed privately, on-device (optional)
• 📝 Saved history — conversations stay in your browser, organized in a sidebar
• ⚡ Streaming replies with a live speed meter, code blocks, and one-click copy
• 📱 Works on mobile browsers on your network, too
• 🪶 No dependencies for the core — the whole server is one short Python file

---

🚀 Get started (about 5 minutes)

1. Install Ollama (the local model engine)

• Mac / Windows / Linux: download from ollama.com/download (or on Mac with Homebrew: brew install ollama)

Then start it:

ollama serve

2. Download a model

ollama pull qwen2.5:7b      # great all-rounder (~4.7 GB)

Low on RAM or want it snappier? Use a smaller one:

ollama pull llama3.2:3b     # fast + light (~2 GB)

Want image understanding? Also grab a vision model:

ollama pull qwen2.5vl:3b    # lets it "see" photos (~3.2 GB)

3. Download this app and run it

git clone https://github.com/ZANYANBU/aether.git
cd aether
python3 chat_server.py

Open http://localhost:8100 in your browser. Pick your model from the dropdown and start chatting. That's it. 🎉

Even simpler: run ./run.sh — it starts Ollama, grabs the models, and opens the app for you.

4. (Optional) Turn on voice

For hands-free 🎧 Live mode and the 🎙️ mic, add local speech-to-text:

python3 -m venv .venv
.venv/bin/pip install faster-whisper
.venv/bin/python chat_server.py

(Talk-back / read-aloud works without this — it uses your browser's built-in voices.)

---

🎛️ Use any model you like

The dropdown at the top automatically lists every model you've installed. Pull whatever you want and it shows up:

ollama pull mistral         # Mistral 7B
ollama pull gemma2:2b       # Google Gemma
ollama pull phi3.5          # Microsoft Phi
ollama pull deepseek-r1:7b  # a reasoning model

Browse the full library at ollama.com/library. When you attach an image, the app automatically switches to a vision-capable model.

---

🧩 How it works

                                              ┌──────────── Ollama ───────────┐
  ┌──────────────┐   POST /chat    ┌────────────────┐   your chosen model     │
  │   Browser    │ ──────────────▶ │  chat_server.py │ ─▶ (or a vision model    │
  │ (index.html) │ ◀────────────── │  (1 python file)│    when you add an image)│
  └──────────────┘  streamed tokens└────────┬───────┘ ◀── streamed tokens      │
     │        ▲                     POST /transcribe    └────────────────────────┘
   🎙️ mic ───┼──── audio ─────────▶┌────────┴────────┐
   🔊 speaker◀─ your system voices  │  local Whisper  │ (optional, on-device)
     (in the browser, fully local)  └─────────────────┘

chat_server.py serves the page and relays your messages to Ollama, streaming the reply back token-by-token. Speech-to-text uses a local Whisper model; text-to-speech uses your browser's built-in voices. Every part runs on your machine.

---

📁 What's inside

File	What it does
chat_server.py	The whole backend: serves the page, lists your models, relays chat, transcribes voice. Pure Python standard library.
index.html	The entire app — chat, model picker, vision, voice, and Live mode — in one file.
run.sh	One-command launcher (starts Ollama, pulls models, opens the app).

---

🔧 Customize

Edit the top of chat_server.py:

DEFAULT_MODEL = "qwen2.5:7b"    # what's selected on first load
VISION_MODEL  = "qwen2.5vl:3b"  # used automatically for images
PORT          = 8100            # change if the port is taken

Voice accuracy: Whisper uses base.en (fast). For better results, change it to small.en in chat_server.py.

---

🩺 Troubleshooting

Problem	Fix
Ollama not reachable	Start it: ollama serve in a terminal.
Dropdown is empty	Pull a model first: ollama pull qwen2.5:7b.
🎙️ mic / 🎧 Live does nothing	Install voice input and run with the venv (setup step 4).
Attaching an image errors	Pull a vision model: ollama pull qwen2.5vl:3b.
Replies are slow	Use a smaller model like llama3.2:3b. Bigger models need more RAM.
Live mode interrupts itself	It's hearing its own voice — use headphones or lower the volume.
address already in use	Change PORT in chat_server.py.

---

🔐 Privacy

This app makes no outbound internet connections while you use it. Your chats are stored only in your own browser (localStorage) and can be deleted anytime. Models, transcription, and speech all run locally. The only time anything is downloaded is the one-time model pull from Ollama during setup.

---

📜 License

MIT — free to use, modify, and share. See LICENSE.

---

If this made local AI click for you, drop a ⭐ — it helps others find it.