AgentX Code

The autonomous AI coding agent built for the terminal.

   /\   /----\  |----\ |\   | ----- \    /     ▄▀▀▀ ▄▀▀█ █▀▀▄ █▀▀▀
  /  \  |    -- |    | | \  |   |    \  /      █    █  █ █  █ █▀▀▀
 /----\ |  ---| |----/ |  \ |   |     \/       ▀▄▄▄ ▀▄▄▀ █▄▄▀ ▀▀▀▀
/      \\-----/ |----/ |   \|   |    /  \

English | 简体中文 | 繁體中文 | 한국어 | Deutsch | Español | Français | Italiano | Dansk | 日本語 | Polski | Русский | Bosanski | العربية | Norsk | Português (Brasil) | ไทย | Türkçe | Українська | বাংলা | Ελληνικά | Tiếng Việt

---

Quick Start

Installation

Install AgentX Code via your preferred toolchain:

via npm (Recommended)

npm install -g @agent-qofeno/agentx-cli@latest

(Also works with bun add -g @agent-qofeno/agentx-cli, pnpm add -g @agent-qofeno/agentx-cli, or yarn global add @agent-qofeno/agentx-cli)

via curl (Linux & macOS)

curl -fsSL https://agentx.js.org/install | bash

via PowerShell (Windows)

irm https://agentx.js.org/install.ps1 | iex

via Homebrew

brew install SohailKhan0525/agentx/agentx

via JSR

bunx jsr add @agent-qofeno/agentx-cli

Launch

Run the agent anywhere inside your workspace:

agentx

Or run directly with an initial instruction:

agentx run "Inspect the project structure and summarize the database schema"

---

Key Features

??? Local Models First (Zero API Keys Required)

Run completely offline with private, sovereign AI. AgentX Code scans your machine, interrogates running runtimes (Ollama, LM Studio, Jan, LocalAI, llama.cpp, GPT4All), detects your CPU, RAM, GPU, and VRAM, and recommends optimal quantized models.

agentx local --scan
agentx local --pull qwen2.5-coder:7b

?? Universal Multi-Model Orchestration

Switch seamlessly between local engines and premier cloud providers:

• Local: Ollama, LM Studio, Jan, llama.cpp, LocalAI
• Cloud: Anthropic (Claude 3.7 Sonnet / Opus), OpenAI (GPT-4o, o1, o3-mini), Google (Gemini 2.5 Pro / Flash), Groq, DeepSeek, Mistral, xAI (Grok), Cerebras, Cohere, Perplexity, OpenRouter, AWS Bedrock, Azure OpenAI

🟡 High-Contrast Yellow & Black Terminal UI

Engineered for intense developer focus with high-performance Solid.js terminal rendering:

• Vivid yellow brand styling (#FFE600 / ANSI yellow with cyan accents)
• Visual multi-file diff inspection and approval
• Split-footer non-intrusive interactive prompt
• Smooth keyboard navigation (Tab switches agents, Ctrl+C cancels, Esc dismisses)

🛠️ Extensible Tool Architecture

• Terminal Execution: Live streaming bash & command executions
• File System: High-speed fuzzy search, AST ripgrep, surgical file patchers
• Web Research: Search and crawl web documentation on demand
• Model Context Protocol (MCP): Connect external tools, databases, and sidecars via stdio or SSE

👥 Subagent System

• build: Primary full-access development agent
• plan: Read-only architectural planning and exploration
• general: Autonomous worker delegation for parallel search and research

---

Local Models Guide

AgentX Code includes an automated hardware profiler that evaluates your machine and matches it to a curated catalog of open-weights models:

Hardware Tier	Recommended Specs	Recommended Models	Performance Profile
Fast	CPU only or <8 GB RAM / 2 GB VRAM	qwen2.5-coder:1.5b llama3.2:3b deepseek-coder:1.3b	Instant auto-completion, lightweight scripts, zero GPU dependency
Balanced	8–16 GB RAM / 4–8 GB VRAM	qwen2.5-coder:7b deepseek-coder-v2:16b codellama:7b	Full refactors, bug fixing, test writing, architectural reasoning
Quality	16+ GB RAM / 8+ GB VRAM	qwen2.5-coder:14b qwen2.5-coder:32b codestral:22b	Complex system designs, multi-file codebases, frontier performance

Local CLI Commands

# Scan system hardware and active local providers
agentx local --scan

# Pull an Ollama model with live progress streaming
agentx local --pull qwen2.5-coder:7b

# Launch interactive local provider setup
agentx local

---

Configuration

AgentX Code configuration files are located in .agentxcode within your project or home directory:

• Project Configuration: <workspace>/.agentxcode/agentx.json
• Global Configuration: ~/.agentxcode/agentx.json

(Full backwards compatibility: Existing .agentx and .opencode directories and configuration files are automatically recognized as fallbacks).

Example agentx.json:

{
  "$schema": "https://agentx.js.org/schema.json",
  "theme": "yellow",
  "provider": {
    "local": {
      "endpoint": "http://127.0.0.1:11434"
    }
  },
  "agent": {
    "build": {
      "model": "local/qwen2.5-coder:7b"
    }
  }
}

---

Migrating from OpenCode

AgentX Code is a standalone fork and evolution of OpenCode:

• Command name: agentx (legacy alias opencode is preserved)
• Package: @agent-qofeno/agentx-cli
• Config folder: .agentxcode/ (with fallback to .agentx/ and .opencode/)
• Pure terminal focus with zero proprietary cloud lock-ins

---

License & Attribution

AgentX Code is licensed under the MIT License.

Based upon the foundational work by the OpenCode authors and contributors.

AgentX Code is maintained and developed by Sohail Khan.