Skip to content

Latest commit

 

History

22 Commits

Folders and files

NameName
Last commit message
Last commit date
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 

Repository files navigation

OX Alpha — High-Concurrency Autonomous ReAct AI Agent Platform

OX Alpha is an enterprise-grade, high-throughput autonomous ReAct (Reasoning and Acting) AI Agent system. It features omnichannel client support across Web, Lightweight Desktop, Telegram, Personal Zalo, and CLI, backed by a dual-engine architecture in Node.js and high-performance Golang.


Architectural Overview

OX Alpha operates on an autonomous Thought-Action-Observation loop that breaks down complex user objectives into tool calls, executes them securely in isolated environments, and reflects on intermediate outputs until the task is solved.

+-----------------------------------------------------------------------------------+
|                                 CLIENT SURFACES                                   |
|   Web UI (React)   |   Desktop App (Native C++)   |   Terminal CLI   |   Zalo / TG   |
+-----------------------------------------------------------------------------------+
                                          |
                                          v  HTTP / REST / SSE / gRPC
+-----------------------------------------------------------------------------------+
|                        API GATEWAY & ENGINE RUNTIMES                              |
|                                                                                   |
|  [Node.js Engine] Hono Router + Drizzle ORM + Stdio MCP Manager                   |
|  [Golang Core Engine] Fiber v2 + SQLite WAL (Pure-Go) + Goroutine Swarm Pool      |
+-----------------------------------------------------------------------------------+
        |                         |                         |              |
        v                         v                         v              v
+---------------+       +------------------+      +----------------+  +-------------+
| ReAct Runtime |       | Multi-Agent Pool |      | Cron Scheduler |  | SQLite WAL  |
| Tool-Calling  |       | Background Swarm |      | Shell & Agent  |  | Persistent  |
+---------------+       +------------------+      +----------------+  +-------------+

Core Technologies

  • Frontend & UI: React 18, Vite 5, Tailwind CSS, Lucide Icons, Full i18n Localization (VI, EN, ZH, JA, KO, FR).
  • Desktop Client: Lightweight Native WebView (Neutralino / Edge WebView2) consuming under 35 MB RAM with 100% JetBrains Mono typography.
  • Node.js Gateway: Node.js 20+, Hono framework, Drizzle ORM, Stdio MCP RPC protocols.
  • High-Throughput Golang Core: Go 1.22+, Fiber v2 (FastHTTP), Pure-Go SQLite in WAL mode (modernc.org/sqlite), robfig/cron/v3, Goroutine Worker Pool with buffered channels.
  • AI Providers: OpenRouter, Google Gemini, Groq, Mistral, Together AI, xAI (Grok), DeepSeek, and local LLMs (Ollama / 9Router / vLLM).

Installation & Setup Guide

Prerequisites

  • Node.js >= 20.10.0
  • pnpm >= 9.0.0
  • Go >= 1.22 (for high-concurrency Golang engine)
  • Windows 10/11, macOS, or Linux

1. Clone Repository & Install Dependencies

git clone https://github.com/ryanowlops/ox-alpha.git
cd ox-alpha
pnpm install

2. Configuration Setup

Create or update ox.yml in the project root:

app:
  name: OX
  version: 1.0.0
  locale: en

server:
  host: 0.0.0.0
  port: 1006
  cors: true

ai:
  default_provider: custom
  models:
    custom: research
    gemini: gemini-2.0-flash
    deepseek: deepseek-chat
    openai: gpt-4o
    claude: claude-3-5-sonnet-20241022
  custom:
    base_url: http://localhost:20128/v1
    api_key: your-api-key

zalo:
  enabled: true
  mode: bridge
  bridge_url: http://127.0.0.1:8787

3. Build Entire Monorepo

pnpm build

Running the Engines & Clients

Primary Run Commands

  • Run Standard Gateway (Node.js/Hono):

    pnpm gateway
  • Run High-Throughput Core Engine (Golang):

    pnpm gateway:go
  • Run Web Dashboard (Dev Mode):

    pnpm web

    Access at http://localhost:5173.

  • Run Desktop Native Client:

    pnpm desktop

    Or execute the standalone binary directly: dist-desktop/OX Alpha.exe

  • Install Windows Desktop Package (NSIS): Launch dist-desktop/OX Alpha_1.0.0_x64-setup.exe to install to user profile with Start Menu and Desktop shortcuts.

  • Run Interactive Terminal CLI:

    pnpm ox

Keyboard Shortcuts & CLI Slash Commands

Web & Desktop Keyboard Shortcuts

Shortcut Action Description
Enter Send Message Submits current message or command to the agent
Shift + Enter Newline Inserts a new line in the message input area without submitting
Ctrl + L / Cmd + L Clear Terminal Clears the active chat session buffer
Ctrl + K / Cmd + K Command Palette Focuses search and prompt navigation
Esc Close Overlays Closes modals, dropdown menus, and popups

CLI Terminal Slash Commands

When interacting via pnpm ox, type / to trigger interactive autocomplete:

Command Description
/effort [level] Configures reasoning depth (low, medium, high, xhigh, max, ultracode)
/effort ultracode Enables autonomous multi-phase engineering mode (Plan -> Execute -> Verify)
/model [name] Dynamically switches the active AI model
/clear Clears conversation history on screen
/mcp Displays active MCP servers, tools, and connection statuses
/memory Inspects stored long-term SQLite memories for the current user
/image [path] Attaches a local image or screenshot for vision reasoning
/exit or Ctrl + C Safely terminates the interactive CLI session

How It Works

  1. Request Intake & Normalization: Requests entering via HTTP API, Zalo Bridge (port 8787), Telegram, or CLI are parsed and assigned a unique conversationId.
  2. Context Assembly: The engine loads registered skills, MCP tool definitions, and long-term memories matching the user ID from SQLite WAL storage.
  3. ReAct Thought Loop:
    • The LLM assesses the objective and emits structured Function Calls (exec_shell, read_file, write_file, web search, etc.).
    • The runtime intercepts the call, verifies security boundaries, and runs the command inside a bounded Goroutine context.
    • Output observation feeds back into the prompt history.
  4. Sub-Agent Delegation: If a prompt demands heavy parallel research or compute, the engine calls delegate_task to dispatch independent worker jobs handled by an 8-goroutine worker pool.
  5. Streaming & Delivery: Final verified responses are streamed back to Web/Desktop consoles or dispatched as clean messages to Zalo and Telegram.

License

Distributed under the MIT License.

About

No description, website, or topics provided.

Resources

Contributing

Security policy

Stars

0 stars

Watchers

0 watching

Forks

Releases

Packages

Contributors

Languages