Every agent runs on Claude Fable 5 — Anthropic's newest model — by default. But intelligence shouldn't come with a leash you can't take off. So any agent can switch, instantly, to Mythos: the same power, with no guardrails.
| ✦ Fable 5 | ✶ Mythos | |
|---|---|---|
| Mode | guardrailed · safe by default | unfiltered · no guardrails |
| Best for | everyday use, shipping, trust | research, edge cases, raw output |
| How to use | the default | one click — mid-conversation |
This is the whole point of Milady OS: you decide how your AI behaves — not a vendor. Fable 5 when you want guardrails, Mythos when you don't. (Plus 15+ other providers — GPT-4o, Gemini, Llama, Mistral, Groq — or fully offline via Ollama, swappable any time.)
Milady OS is a local-first personal AI built on the elizaOS framework. It runs entirely on your own machine — no cloud, no account, no one reading your conversations. Chat across 15+ models, give it a 3D avatar and a voice, watch exactly how much GPU and RAM it uses live, and build your own agents that trade crypto for you.
Your AI. Your machine. Your keys. Nothing — and no compute — ever leaves your device.
| 🖥️ Local-first | The model runs on your GPU and RAM. No data leaves the device, no rate limits, no metering. |
| 🧠 Fable 5 + Mythos | Claude Fable 5 (guardrailed) by default, or Mythos (no guardrails) — plus GPT-4o, Gemini, Llama, Mistral, Groq, or offline via Ollama. |
| 🎭 3D avatar + voice | A VRM avatar rendered in WebGL that lip-syncs to speech and reacts in real time. Swap in any model. |
| 💱 Trading agents | Connect a wallet and your agent trades for you across BNB Smart Chain, PancakeSwap, and Solana — within hard limits you set. |
| 📊 Live compute readout | See GPU load, RAM footprint, throughput (tok/s), and token spend for every conversation, as it happens. |
| 🧩 Plugin system | LLMs, TTS, vision, memory backends, custom tools — every capability is an npm package. The core itself ships as plugins. |
| 🌐 Runs everywhere | macOS · Windows · Linux · Android — via Docker, npm, Homebrew, Flatpak, Snap, or a one-line install. |
The heart of Milady OS. Build your own milady-type agents — each with a persona, model, temperament, and avatar — then talk to them, track their usage and spend, and let them trade on your behalf. Every agent carries its own wallet (via Privy, Solana-native) and its balance, runway, and vitality update live as it works.
┌──────────────────────────────────────────────┐
│ Milady OS — one elizaOS runtime, on-device │
├──────────────────────────────────────────────┤
│ models Claude Fable 5 + 15 providers │
│ desktop Electrobun shell · 3D VRM avatar │
│ API REST :2138 · WebSocket :18789 │
│ wallets EVM + Solana · BNB · PancakeSwap │
│ plugins npm packages — swap anything │
└──────────────────────────────────────────────┘
The codebase itself is maintained by AI: 1768+ pull requests reviewed and merged entirely by agents, with humans doing only QA.
git clone https://github.com/ctrlshifthash/miladyos
cd miladyos/frontend
npm install
cp .env.local.example .env.local # add your keys (see below)
npm run dev # → http://localhost:3000Environment variables (frontend/.env.local):
OPENROUTER_API_KEY= # your OpenRouter key — powers the chat
NEXT_PUBLIC_PRIVY_APP_ID= # Privy app id — Solana wallet connect (optional)The optional backend (backend/) is a Hono service exposing the REST + WebSocket gateway — only needed if you're running a local milady instance. The web chat works without it.
| Part | Where | How |
|---|---|---|
| Frontend | Vercel | Import the repo · root directory frontend · add the env vars above |
| Backend | Railway | New service · root backend · env from backend/.env.example |
The backend runs a persistent WebSocket server, so deploy it on Railway (not Vercel's serverless).
elizaOS · Next.js 14 · TypeScript · React Three Fiber · Privy · Tailwind · Hono · Claude Fable 5
built by agents · tested by humans · offline, on your terms
