milady-ai/milady

terminally online

Milady – a privacy‑first, locally‑run AI companion

What it is – Milady is a personal AI assistant that runs on your own machine by default. It is built on the open‑source elizaOS stack and can also hook into the hosted Eliza Cloud or any self‑hosted Milady backend you run elsewhere. The goal is a “AI girlfriend/assistant” that feels responsive, looks cute (3‑D avatar), and never phones home unless you explicitly enable a remote backend.

Key capabilities

  • Local‑first inference – the core agent runs on your CPU/GPU; you can pick a model provider (Anthropic Claude, OpenAI GPT, or a local Ollama model) during onboarding.
  • Multi‑platform front‑ends – a desktop app (macOS, Windows, Linux) with a WebGL‑based 3‑D avatar, plus Telegram/Discord connectors and a browser‑based dashboard UI.
  • Energy‑aware desktop app – the UI throttles rendering when the window is hidden or the laptop is on battery, aiming for low GPU usage.
  • Action streaming – long‑running tool calls (e.g., web searches, file ops) are streamed back to the UI as progressive updates, replacing the previous line rather than appending a long log.
  • DeFi integration – built‑in support for BNB Smart Chain (PancakeSwap trades, wallet generation, meme‑token discovery via the Binance Skills Hub). Private keys can be supplied locally or managed by the cloud backend.
  • Plugin system – a registry of “skills” (e.g., the meme‑rush skill) can be installed from the dashboard or via the CLI.
  • Remote‑backend mode – you can run the Milady backend on a VPS, homelab, or Tailscale‑exposed server and have the desktop/web UI talk to it securely with an API token.

How you get it

Platform Install method
macOS (Apple Silicon & Intel) Download the signed DMG from the Releases page
Windows Download the installer EXE from Releases
Linux (x86_64) Download the tar.gz installer from Releases and run ./installer
Android Google Play store or direct APK from Releases
iOS Coming soon (App Store)
CLI‑only npm install -g miladyai or the one‑liner `curl -fsSL https://get.milady.ai

Typical first‑run flow

  1. Run milady setup (or the installer).
  2. An interactive wizard asks you to pick a persona (e.g., Sakuya, Reimu) and a voice style.
  3. Choose a language model – you can connect to Anthropic, OpenAI, or a local Ollama model.
  4. (Optional) Provide an EVM private key if you want BNB‑Chain trading.
  5. The dashboard becomes available at http://localhost:2138; a TUI appears in the terminal showing live activity.

Command‑line overview

milady                 # start interactive UI + dashboard (default)
milady start           # run only the API server (no TUI)
milady dashboard       # open the web UI in a browser
milady plugins install <name>   # add a skill (e.g., meme‑rush)
milady config get <key>         # read a config value
milady doctor          # run a health‑check

Remote deployment – If you set MILADY_API_BIND=0.0.0.0 and generate a strong MILADY_API_TOKEN, the backend can be reached from other machines. Pair it with a Tailscale tunnel or expose it behind HTTPS, then point the desktop or web UI to the remote address during onboarding.

Security notes

  • By default the API binds only to 127.0.0.1; no external access is possible.
  • When you expose the API, you must set MILADY_API_TOKEN and optionally MILADY_ALLOWED_ORIGINS to restrict who can call it.
  • The desktop app does not upload conversation data unless you enable the Eliza Cloud backend.

Why it matters – Milady demonstrates a full‑stack, privacy‑centric AI assistant that combines local LLM inference, a plug‑in‑driven tool system, and optional blockchain interaction—all wrapped in a consumer‑friendly desktop experience. It is a concrete example of the “local‑first AI agent” paradigm that many developers are exploring as an alternative to always‑online SaaS chatbots.

Related

  • Project
    cosmicstack-labs/mercury-agent

    Mercury is a locally‑run, permission‑hardened AI assistant that can chat via CLI, web UI, or Telegram, remembers you with a SQLite‑backed “second brain”, enforces daily token budgets, and is extensible through community‑contributed skills. Install with a one‑line script (no Node required) or via npm, then run `mercury up` to keep it alive as a background service.

  • Project
    theJayTea/WritingTools

    Writing Tools is a free, open‑source desktop app for Windows, Linux and macOS that lets you invoke an LLM anywhere on your computer with a hot‑key. It can proofread, rewrite, change tone, translate, summarise web pages or YouTube transcripts, and even chat with the model. You can use cloud APIs (Gemini, OpenAI, Anthropic, etc.) or run local models via Ollama, llama.cpp or Apple‑silicon MLX. The app is privacy‑focused (no telemetry, local storage of API keys) and licensed under GPL‑v3.

  • Project
    lemonade-sdk/lemonade

    Lemonade is a community‑driven local AI server (and embeddable binary) that runs LLM, speech, TTS, and image models on CPUs, GPUs, or AMD NPUs. It mimics OpenAI‑style APIs, includes a model manager, hardware‑auto‑optimizing back‑ends, and a marketplace of ready‑made integrations.

  • Project
    OpenMinis/OpenMinis

    OpenMinis is an open‑source mobile AI agent (iOS + Android) that lets you run any LLM (Claude, GPT, Gemini, etc.) on‑device. It ships a sandboxed Alpine Linux shell (iSH on iOS, PRoot on Android) so the agent can install packages, run scripts, and manipulate files, while also exposing native device services (Health, Calendar, HomeKit, Bluetooth, clipboard, media, etc.) as tools. Users create reusable “skills” (folders with a `SKILL.md` and optional scripts) and organise work into separate workspaces. Typical uses include nutrition logging, morning briefings, task extraction from chats, syncing with Obsidian, and share‑to‑event automation. The codebase is Swift/SwiftUI for iOS, Kotlin/Compose for Android, and is GPL‑v3 licensed.

  • Project
    vellum-ai/vellum-assistant

    A personal AI assistant with eight types of memory that learns user behavior and takes proactive actions across multiple apps and communication channels.