liruifengv / amon-agent

Amon Coworker Your AI coworker running on your desktop 中文 | English About Amon Amon is an intelligent AI coworker that runs locally on your desktop. It doesn't just chat with you — it actually helps you get work done: writing code, executing commands, searching for information, and managing files. Amon features a custom three-layer agent architecture with a provider-agnostic AI streaming layer, supporting multiple LLM providers including Anthropic Claude, OpenAI, Google Gemini, and API-compatible providers out of the box. Starting with , Amon no longer depends on Claude Agent SDK. The runtime is implemented in-repo with a provider-agnostic AI layer, a framework-agnostic agent core, and Electron-specific integration. Breaking Changes in 0.3.0 • Claude Agent SDK has been removed and replaced by Amon's own agent core and runtime. • Settings and provider configuration moved to the new / / schema. • Existing settings are migrated on a best-effort basis. Older provider-specific fields and deprecated options may need to be reconfigured manually after upgrading. Feature Overview Let's take a look at Amon's features through screenshots. Amon can think through your messages, execute tool calls, and complete your tasks. Amon supports both dark and light theme modes. Amon can display file modification diffs. Amon supports sending image messages. Amon allows you to add multiple API providers with built-in support for Anthropic Claude, OpenAI, Google Gemini, and API-compatible providers (GLM, MiniMax, Kimi, etc.). Amon works on a per-workspace (folder) basis. You can set up multiple workspaces. Default workspace: Amon supports Agent Skills — you can install Skills to add specialized capabilities to Amon. Getting Started Installation Visit the Releases page to download the installer for your platform. Note for macOS Users Since the app is not Apple-signed, macOS may block it from running. After downloading, run the following command in the terminal to remove the quarantine attribute: Configuration After first launch, follow these steps to configure: • **Configure AI Provider** Go to → , create and enable the AI provider you want to use • **Create a Workspace** Go to → , create a new workspace and select a local folder as the project root Default workspace: • **Start Using** Return to the main screen, click , select a workspace and start chatting Development Guide Prerequisites • Node.js 18+ or Bun 1.0+ • macOS / Windows / Linux Development Commands Build and Package Project Structure Architecture Amon adopts a three-layer agent architecture, with each layer cleanly decoupled: • **AI Layer** ( ) — Provider-agnostic streaming abstraction. Global provider registry with 4 built-in providers. Normalizes all responses into a unified stream. • **Agent Layer** ( ) — Framework-agnostic class. Dual-loop architecture: inner loop (LLM call -> tool execution -> steering check), outer loop (follow-up queue -> repeat). Tool input validated with Zod schemas. • **Integration Layer** ( ) — Wires Agent into Electron. resolves providers, models, skills, and workspace bootstrap files per session. bridges agent events to session store mutations and push notifications. Tech Stack **Core Frameworks** • Electron — Cross-platform desktop apps • React 19 — UI framework • TypeScript — Type safety **AI Layer** • Custom provider-agnostic streaming layer • Anthropic SDK / OpenAI SDK / Google GenAI SDK • Dual-loop agent architecture (tool execution + follow-up) **Frontend** • Tailwind CSS + Shadcn/ui — UI design • Zustand — State management • Streamdown — Streaming Markdown rendering **Build Tools** • Vite — Lightning-fast builds • Electron Forge — Packaging and distribution • Bun — Runtime and package manager **Data & Validation** • Zod v4 — Runtime type validation and tool input schemas • Shiki — Code syntax highlighting • i18next — Internationalization (en, zh) License This project is open-sourced under the AGPL-3.0 license. --- Built with ❤️ by @liruifengv