AI Models & Vai trò

Galaxy CLI hỗ trợ nhiều AI models khác nhau, cho phép bạn linh hoạt chọn agent phù hợp với nhu cầu.

Các AI Models được hỗ trợ

Google Gemini

Models: gemini-2.5-flash, gemini-pro

Orchestration, tool calling, multi-language support

Anthropic Claude

Models: claude-3.5-sonnet, claude-3-opus

Complex reasoning, code generation, analysis

OpenAI Codex

Models: gpt-4, gpt-3.5-turbo

Code completion, documentation, refactoring

Ollama (Local)

Models: qwen3-coder, gpt-oss, codellama

Local inference, privacy-focused, customizable

Cấu trúc Agents

Galaxy CLI sử dụng kiến trúc orchestrator với các specialized agents:

                    ┌──────────────────────┐
                    │  User Input (CLI)    │
                    └──────────┬───────────┘
                               │
                               ▼
          ┌────────────────────────────────────┐
          │   Orchestrator Agent               │
          │   (Gemini / Claude / GPT-4)        │
          │   - Tool calling                   │
          │   - Workflow decisions             │
          │   - Multi-language responses       │
          └────────────┬───────────────────────┘
                       │
      ┌────────────────┼────────────────┐
      │                │                │
      ▼                ▼                ▼
┌───────────┐   ┌───────────┐   ┌────────────┐
│ Analysis  │   │ Planning  │   │ Code Gen   │
│ Agent     │   │ Agent     │   │ Agent      │
│ (Any AI)  │   │ (Any AI)  │   │ (Any AI)   │
└───────────┘   └───────────┘   └────────────┘

1. Orchestrator Agent

Supported Models:

Google Gemini 2.5 Flash (default)
Claude 3.5 Sonnet
GPT-4

Vai trò:

Central coordinator & workflow manager
Tool calling & function execution
Context management
Multi-language support (VI/EN)

Capabilities:

// Tool calling
orchestrator.callTool('file_read', { path: './app.tsx' });

// Agent coordination
const analysis = await orchestrator.callTool('analyze', {...});
const plan = await orchestrator.callTool('plan', {...});

// Execution
for (const step of plan.steps) {
  await orchestrator.executeStep(step);
}

2. Analysis Agent

Supported Models:

Claude 3.5 Sonnet
GPT-4
Ollama: gpt-oss:120b

Vai trò:

Business & requirements analysis
Tech stack recommendations
Feature breakdown
Data modeling

Output Example:

{
  projectName: "Next.js E-commerce",
  type: "create_project",
  coreFeatures: [
    { name: "Product Catalog", priority: "must-have" },
    { name: "Shopping Cart", priority: "must-have" },
    { name: "Checkout", priority: "must-have" }
  ],
  technicalStack: {
    frontend: ["Next.js 14", "TypeScript", "Tailwind"],
    backend: ["Prisma", "PostgreSQL"]
  }
}

3. Planning Agent

Supported Models:

Claude 3.5 Sonnet
Ollama: qwen3-coder:480b
GPT-4

Vai trò:

Feature-driven planning
Tool selection
Execution strategy
Time estimation

Planning Strategy:

[
  { step: 1, tool: 'command_run', action: 'Initialize project' },
  { step: 2, tool: 'install_dependencies' },
  { step: 3, tool: 'code_generate', featureName: 'Feature A' },
  { step: 4, tool: 'code_generate', featureName: 'Feature B' },
  { step: 5, tool: 'test_generate' },  // if testEnabled
  { step: 6, tool: 'code_review' }     // if reviewEnabled
]

4. Code Generation Agent

Supported Models:

Claude 3.5 Sonnet (recommended)
Ollama: qwen3-coder:480b
GPT-4
Codex

Vai trò:

Generate production-ready code
File management
Self-contained execution
Result reporting

Integrated Tools:

file_write - Write code files
file_read - Read existing code
command_run - Run commands
file_search - Search codebase

Chọn AI Model

Mode tự động (Recommended)

# Galaxy tự động chọn agent phù hợp
galaxy

Galaxy sẽ:

Sử dụng Gemini cho orchestration (nhanh, rẻ)
Sử dụng Claude cho code generation (chất lượng cao)
Sử dụng local models khi có thể (privacy)

Mode thủ công

# Chỉ định agent cụ thể
galaxy --agent gemini
galaxy --agent claude
galaxy --agent gpt4
galaxy --agent ollama:qwen3-coder

Cấu hình trong file

{
  "agent": {
    "orchestrator": "gemini",
    "analysis": "claude",
    "planning": "claude",
    "codeGen": "claude"
  }
}

So sánh Models

Model	Provider	Speed	Quality	Cost	Use Case
Gemini 2.5 Flash	Google	⚡⚡⚡	⭐⭐⭐	💰	Orchestration
Claude 3.5 Sonnet	Anthropic	⚡⚡	⭐⭐⭐⭐⭐	💰💰	Code Gen, Analysis
GPT-4	OpenAI	⚡⚡	⭐⭐⭐⭐	💰💰💰	Complex reasoning
Ollama (Local)	Local	⚡	⭐⭐⭐	🆓 Free	Privacy, Offline

Configuration

API Keys

# .env hoặc ~/.galaxyrc
GEMINI_API_KEY=your_key_here
ANTHROPIC_API_KEY=your_key_here
OPENAI_API_KEY=your_key_here

Model Parameters

// Adjust creativity/determinism
{
  "modelConfig": {
    "temperature": 0.7,  // 0-1, higher = more creative
    "maxTokens": 8192,
    "topP": 0.95
  }
}

Best Practices

Các bước tiếp theo

Architecture - Hiểu kiến trúc tổng thể
Tools - Tools có sẵn cho agents
Configuration - Cấu hình chi tiết