One prompt. One agent. One complete website.
AgentForge is a conversational CLI agent that autonomously thinks, plans, and builds production-quality websites — all from a single text prompt in your terminal.
▶ You: Clone the Scaler Academy website
🚀 [START] [Step 1] Understanding your request...
│ I will build a high-fidelity clone of the Scaler Academy website...
🧠 [THINK] [Step 2] Reasoning...
│ Let me plan the file structure. I'll need styles.css, index.html, script.js...
🔧 [TOOL] [Step 3] Executing action...
│ → createFile("output/styles.css") ✓ (6.4 KB)
🔧 [TOOL] [Step 5] Executing action...
│ → createFile("output/index.html") ✓ (8.2 KB)
🔧 [TOOL] [Step 7] Executing action...
│ → createFile("output/script.js") ✓ (3.1 KB)
🔧 [TOOL] [Step 9] Executing action...
│ → openInBrowser("output/index.html") ✓ Opened in browser
╭──────────────────── OUTPUT ────────────────────╮
│ ✅ Task Complete! │
│ Built: Header, Hero, Highlights, Curriculum, │
│ Footer — all responsive with animations. │
╰─────────────────────────────────────────────────╯
╭──────────────── STATS ─────────────────╮
│ 📊 Session Summary │
│ ⏱ Time Elapsed: 45.2s │
│ 🔄 Steps Taken: 12 │
│ 📁 Files Created: 3 │
│ • output/styles.css (6.4 KB) │
│ • output/index.html (8.2 KB) │
│ • output/script.js (3.1 KB) │
╰────────────────────────────────────────╯
|
The agent doesn't just generate code — it thinks step by step. Each response follows a structured |
Supports multiple API keys with automatic rotation. The agent cycles through your key pool on every request, maximizing throughput and minimizing rate limit hits. |
|
10 automatic retries with exponential backoff (5s → 30s). The agent never crashes from rate limits — it patiently waits and retries with a visual progress indicator. |
After every task, see a detailed summary: time elapsed, steps taken, files created with sizes. Know exactly what the agent built and how long it took. |
|
Gradient ASCII art, color-coded reasoning steps, animated spinners, boxed outputs — this isn't your average CLI tool. It's a visual experience. |
Only 3 whitelisted tools: |
┌──────────────────────────────────────────────┐
│ AgentForge CLI v2.0 │
│ Built by Abdul Kalam Azad │
└──────────────────┬───────────────────────────┘
│
┌──────────────────▼───────────────────────────┐
│ index.js (Agent Core) │
│ │
│ ┌─────────────┐ ┌────────────────────┐ │
│ │ Readline │ │ Round-Robin Key │ │
│ │ Interface │ │ Load Balancer │ │
│ └──────┬──────┘ └────────┬───────────┘ │
│ │ │ │
│ ┌──────▼──────────────────▼───────────┐ │
│ │ ReAct Reasoning Loop │ │
│ │ │ │
│ │ START ──▶ THINK ──▶ TOOL ──▶ OUTPUT │ │
│ │ ▲ │ │ │
│ │ └─ OBSERVE ┘ │ │
│ └──────────────────┬──────────────────┘ │
└─────────────────────┼───────────────────────┘
│
┌───────────────────────────┼───────────────────────────┐
│ │ │
┌─────────▼─────────┐ ┌───────────▼──────────┐ ┌──────────▼──────────┐
│ prompts.js │ │ tools.js │ │ ui.js │
│ │ │ │ │ │
│ System Prompt │ │ createFile() │ │ Welcome Banner │
│ Scaler Design │ │ createDirectory() │ │ Step Display │
│ Knowledge Base │ │ openInBrowser() │ │ Session Stats │
│ Tool Descriptions │ │ executeTool() │ │ Gradients/Spinners │
└────────────────────┘ └───────────────────────┘ └─────────────────────┘
graph LR
A[👤 User Input] --> B[🚀 START]
B --> C[🧠 THINK]
C --> D[🔧 TOOL]
D --> E[👁️ OBSERVE]
E --> C
C --> F[✅ OUTPUT]
F --> G[📊 Stats]
style A fill:#6C63FF,color:#fff,stroke:none
style B fill:#3B82F6,color:#fff,stroke:none
style C fill:#06B6D4,color:#fff,stroke:none
style D fill:#10B981,color:#fff,stroke:none
style E fill:#F59E0B,color:#fff,stroke:none
style F fill:#22C55E,color:#fff,stroke:none
style G fill:#38BDF8,color:#fff,stroke:none
| Step | Icon | Purpose | What You See |
|---|---|---|---|
| START | 🚀 | Acknowledge & plan | Blue banner with task summary |
| THINK | 🧠 | Reason about next action | Cyan italic reasoning text |
| TOOL | 🔧 | Execute a file operation | Green tool name + arguments |
| OBSERVE | 👁️ | Process tool result | Amber result feedback |
| OUTPUT | ✅ | Deliver final result | Green boxed completion message |
| STATS | 📊 | Show session metrics | Blue stats summary box |
The agent has access to 3 purpose-built, safe tools:
// 📄 Create a file with content
createFile("output/index.html", "<html>...</html>")
// → "File created successfully: output/index.html (8.2 KB)"
// 📁 Create a directory
createDirectory("output/assets")
// → "Directory created successfully: output/assets"
// 🌐 Open in browser
openInBrowser("output/index.html")
// → "Opened output/index.html in the default browser successfully"💡 Why only 3 tools? Instead of giving the agent a dangerous
executeCommandtool that could run anything, we provide specific, safe tools. This makes the agent reliable and prevents accidental system damage.
| Requirement | Version |
|---|---|
| Node.js | 18+ |
| npm | 9+ |
| Gemini API Key | Get one free → |
# Clone the repository
git clone https://github.com/abdulkalamazad1001/agentforge-cli.git
# Navigate to the project
cd agentforge-cli
# Install dependencies
npm install
# Configure your API key(s)
echo "GEMINI_API_KEYS=your-key-here" > .env
# For multiple keys (Round-Robin load balancing):
echo "GEMINI_API_KEYS=key1,key2,key3" > .env
# Launch AgentForge
npm start| Command | Action |
|---|---|
Clone the Scaler Academy website |
Build a full website clone |
Build a portfolio with dark mode |
Create a custom website |
help |
Show available commands |
clear |
Clear the terminal |
exit / quit / q |
Exit AgentForge |
When asked to clone the Scaler Academy website, the agent generates:
| Section | Features |
|---|---|
| 📌 Top Contact Bar | Dark gradient with phone number and CTA |
| 🧭 Sticky Header | White navbar with logo, nav links, mobile hamburger |
| 🦸 Hero Section | Animated gradient, floating orbs, glassmorphic form |
| 📋 Course Overview | 3 informational cards with shadow effects |
| ⭐ Key Highlights | 6 colorful animated cards with hover effects |
| 💼 Advisor CTA | Dark section with benefits checklist |
| 📚 Curriculum | Interactive tab switcher (Beginner/Intermediate/Advanced) |
| 🔗 Footer | Dark footer with links, socials, copyright |
Design Quality: Responsive • CSS Animations • Google Fonts • Pixel-perfect colors
| Technology | Purpose | Why |
|---|---|---|
 |
Runtime | Fast, async, perfect for CLI tools |
 |
AI Engine | Latest reasoning model with free tier |
 |
Terminal Colors | Rich, 256-color terminal output |
 |
Spinners | Smooth animated loading indicators |
 |
Boxes | Styled terminal boxes for output |
 |
Gradients | Multi-color gradient text effects |
 |
Config | Secure environment variable management |
AgentForge CLI/
│
├── 📄 index.js # Agent core — ReAct loop, Round-Robin balancer, retry logic
│ # ~400 lines │ @author Abdul Kalam Azad
│
├── 📄 prompts.js # System prompt — Scaler design knowledge, tool descriptions
│ # ~180 lines │ @author Abdul Kalam Azad
│
├── 📄 tools.js # Safe toolset — createFile, createDirectory, openInBrowser
│ # ~190 lines │ @author Abdul Kalam Azad
│
├── 📄 ui.js # Terminal UI — gradients, spinners, stats, step display
│ # ~380 lines │ @author Abdul Kalam Azad
│
├── 📄 package.json # Dependencies and scripts
├── 📄 .env # API keys (git-ignored)
├── 📄 .gitignore # Excludes .env, node_modules, output/
└── 📄 README.md # You are here
🔁 Round-Robin API Key Pool — Click to expand
// Supports N keys via comma-separated env var
const apiKeys = process.env.GEMINI_API_KEYS.split(",");
let currentKeyIndex = 0;
// Every API call rotates to the next key
const activeKey = apiKeys[currentKeyIndex];
currentKeyIndex = (currentKeyIndex + 1) % apiKeys.length;- Each key gets its own
GoogleGenerativeAIclient instance - Global conversation history persists across key switches
- Zero context loss during rotation
🛡️ Exponential Backoff Retry — Click to expand
Attempt 1 → Wait 5s [█░░░░░░░░░]
Attempt 2 → Wait 10s [██░░░░░░░░]
Attempt 3 → Wait 15s [███░░░░░░░]
Attempt 4 → Wait 20s [████░░░░░░]
Attempt 5 → Wait 25s [█████░░░░░]
Attempt 6+ → Wait 30s [██████░░░░] (capped)
- 10 retries cover a full ~2.5 minute rate limit window
- Auth errors (401/403) fail fast — no wasted retries
- Visual progress bar shown to user during waits
2-3 minute walkthrough showing the agent running live and the generated website
This project is licensed under the MIT License — see the LICENSE file for details.
Built with ❤️ by Abdul Kalam Azad
