Adame

most recent version downloadable in https://adame.ai

Adame is an agentic desktop app for macOS. It pairs a native Electron + React interface with a local Python agent backend, so you can run multi-step AI agents — with tools, skills, an embedded terminal, and your own API keys — directly on your machine.

Features

• Bring your own model. Works with OpenAI, Anthropic, DeepSeek, or a local provider. API keys are stored encrypted on-device via the OS keychain (Electron safeStorage) — never in plaintext, never sent anywhere but the provider you configure.
• Sign in with ChatGPT. Optional OpenAI OAuth as an alternative to an API key.
• Skills. A large, extensible library of reusable agent skills (skills/) covering creative work, research, productivity, dev tooling, and more.
• Tools & MCP. Built-in tools (filesystem, terminal, and more) plus Model Context Protocol client support for connecting external tool servers.
• Embedded terminal. A real pty-backed terminal (xterm.js) inside the app.
• Durable runs. Turn checkpoints, context compaction, and on-device memory keep long agent sessions coherent.
• image Generation in swarms Tired of ChatGPT working on one image at a time per session? Adame gets you as many as you wish at once.

Architecture

Adame is two processes that talk over JSON-RPC on stdio:

• src/main — Electron main process: window, IPC, provider credentials, updater, and the bridge that spawns and supervises the backend.
• src/preload — the typed, sandboxed bridge exposed to the renderer.
• src/renderer — the React UI.
• src/shared — types shared across processes.
• backend/adame_backend — the agent runtime: providers, tools, skills loader, permissions, checkpoints, memory, MCP client, and more.

In development the backend runs straight from Python source. In a packaged build it is compiled to a standalone binary (PyInstaller) and bundled, so end users don't need Python installed.

Getting started

Prerequisites

• macOS (Adame is currently macOS-focused)
• Node.js 20.19+ (required by Vite 7) and npm
• Python 3.11 or newer (for development)

Install

npm install

# Backend Python dependencies (into a virtualenv is recommended)
python3 -m venv .venv
source .venv/bin/activate
pip install -e backend

Run in development

npm run dev

This launches the Electron app with hot-reload. The backend is started automatically from backend/server.py.

If you use a virtualenv or a non-default interpreter, point Adame at it with the ADAME_PYTHON environment variable:

hljs language-bash

Copy

ADAME_PYTHON="$(pwd)/.venv/bin/python" npm run dev

Configure a provider

On first launch, open Settings → Account and either add an API key (OpenAI / Anthropic / DeepSeek) or sign in with ChatGPT. Keys are encrypted and stored locally.

Building a macOS app

# Compile the backend to a standalone binary, then build + package the app
npm run package:mac

Note: signing, notarization, and auto-update publishing are configured for the original project's infrastructure. If you build your own distributable, update the build config in package.json (signing identity, appId, and the publish URL) to your own.

Configuration reference

Project layout

.
├── src/            # Electron app (main / preload / renderer / shared)
├── backend/        # Python agent backend
│   ├── server.py           # JSON-RPC stdio entry point
│   └── adame_backend/      # runtime: providers, tools, skills, ...
├── skills/         # bundled agent skills
├── scripts/        # build tooling
└── build/          # app icon + macOS entitlements

Contributing

Issues and pull requests are welcome. Please keep changes focused and, where it makes sense, verify a dev build runs (npm run dev) before opening a PR.

License

License to be determined.

Some bundled skills under skills/ originate from third parties and carry their own license files — see the individual skill directories.