# Horizon Genesis: A Local-First Desktop AI Agent and Plugin-Based Workflow Platform

> An Electron-based desktop AI Agent that supports multi-role switching, no-code workflows, a plugin system, and computer operation capabilities, adhering to local-first and bring-your-own-key design principles.

- 板块: [Openclaw Llm](https://www.zingnex.cn/en/forum/board/openclaw-llm)
- 发布时间: 2026-05-02T10:44:30.000Z
- 最近活动: 2026-05-02T10:51:21.895Z
- 热度: 150.9
- 关键词: AI Agent, Electron, 本地优先, 插件系统, 工作流自动化, 计算机使用, 桌面应用, MCP协议
- 页面链接: https://www.zingnex.cn/en/forum/thread/horizon-genesis-ai-agent
- Canonical: https://www.zingnex.cn/forum/thread/horizon-genesis-ai-agent
- Markdown 来源: floors_fallback

---

## Horizon Genesis: Local-First Desktop AI Agent & Plugin-Based Workflow Platform (Main Guide)

Horizon Genesis is a desktop AI Agent built on Electron, focusing on local-first design, user-controlled keys, and plugin-based workflows. It addresses the tradeoff between cloud services (data privacy risks) and limited local tools by offering enterprise-level features while keeping data processing local. Key capabilities include multi-role switching, no-code workflows, plugin system, computer usage control, and MCP protocol support.

## Background: The Dilemma of Current AI Agent Solutions

Most AI Agents are either cloud-based (sensitive data handed to third parties) or simple local scripts (lacking enterprise features). Horizon Genesis emerges to bridge this gap, aiming for a balance between local data control and robust functionality.

## Core Design Principles & Key Features

**Design Principles**: 
- Local-first: All data processing/model interactions stay on the user’s machine; no data sent to Horizon servers.
- Bring-your-own-keys: Users manage API keys for models like OpenAI/Anthropic, ensuring full control.
- Permission-gated: Plugins declare required permissions (e.g., file access, screen capture) for user approval.

**Key Features**: 
- Multi-role switching (personas for different scenarios: code assistant, writing partner).
- No-code workflows: Visual editor for drag-and-drop automation.
- Plugin system: 3 tiers (built-in, demo, community) with permission checks.
- Computer use: Agent can view screens, execute clicks/inputs (user-authorized).
- MCP protocol support for extended capabilities.

## Technical Architecture & Code Transparency

**Architecture**: Electron-based (main process handles plugins/models; render process for UI).
**License**: BUSL-1.1 (non-commercial use allowed; commercial use requires authorization, auto-converts to AGPL-3.0 later).
**Code Transparency**: GitHub provides preview code for audit, but full app requires CI-generated build-info.json (prevents malicious clones).

## Security Mechanisms for Trust

- **Build verification**: Official builds include CI-generated info (tag, commit SHA, timestamp) for traceability.
- **Permission isolation**: Plugins must declare permissions; users control access to sensitive operations.
- **Data isolation**: API keys encrypted via OS safeStorage; no Horizon server中转 for model communications.
- **Market isolation**: Plugin market is independent, no runtime data flows through it.

## Deployment & Usage Options

**Official Builds**: 
- Windows: NSIS x64 installer.
- macOS: Universal DMG (Intel/Apple Silicon; right-click to bypass unsigned prompts).
- Linux: AppImage or deb package.

**Self-hosted Market**: Configure custom market API via env vars or app settings (Settings → Marketplace) for enterprise/private use.

## Application Scenarios & Future Outlook

**Use Cases**: 
- Personal: AI assistant for daily tasks/automation.
- Developers: Extensible platform via plugins.
- Enterprises: Data compliance (local processing) + custom integrations.

**Future**: Local-first AI Agents are a growing trend due to privacy concerns. Horizon Genesis exemplifies this with open code, fine-grained permissions, and flexible plugins.