CLI Build

A versatile CLI tool built with Bun and TypeScript that provides AI-powered modes for code assistance, planning, and Telegram integration.

🚀 Features

• CLI Mode — Agent, Plan, and Ask sub-modes for code assistance
• Telegram Mode — Interact with the AI through a Telegram bot
• AI-Powered — Uses NVIDIA AI models through the NVIDIA API
• Web Search — Optional Firecrawl integration for web research
• Safe Operations — Changes can be staged and require approval before being applied
• Tool Calling — AI agents can inspect and work with the project through configured tools
• Plan Mode — Research a codebase and generate implementation plans

📋 Prerequisites

• Bun installed
• NVIDIA API key — required
• Firecrawl API key — optional, for web search
• Telegram Bot Token — optional, for Telegram mode

🔧 Setup

1. Install dependencies

bun install

2. Link the CLI

To use cli-build as a global command from any directory, link the project using Bun:

bun link

3. Configure through CLI

After installing dependencies, configure the CLI using the following commands:

# NVIDIA configuration
cli-build config set nvidia-api-key "your_nvidia_api_key_here"
cli-build config set model "nvidia/nemotron-3-super-120b-a12b"

# Fallback model if the primary model is unavailable
cli-build config set model "nvidia/nemotron-3-ultra-550b-a55b"

# Firecrawl - Web Search
cli-build config set firecrawl-api-key "your_firecrawl_api_key_here"

# Telegram Mode
cli-build config set telegram-bot-token "your_telegram_bot_token_here"
cli-build config set telegram-owner-id "your_telegram_user_id_here"

4. Configure environment variables in .env file (optional)

Create a .env file in the project root.

You can use .env.example as a template:

# Required
NVIDIA_API_KEY=your_nvidia_api_key_here

# Optional but recommended
FIRECRAWL_API_KEY=your_firecrawl_api_key_here

# Optional - for Telegram mode
TELEGRAM_BOT_TOKEN=your_telegram_bot_token_here
TELEGRAM_OWNER_ID=your_telegram_user_id_here

# NVIDIA Models
NVIDIA_DEFAULT_MODEL=nvidia/nemotron-3-super-120b-a12b

If the primary model is unavailable, you can change the value to:

NVIDIA_DEFAULT_MODEL=nvidia/nemotron-3-ultra-550b-a55b

If a value is configured in both CLI configuration and .env, CLI configuration takes priority.

Keep your real API keys and tokens private. Never commit .env or ~/.cli-build/config.json to Git.

5. Run the application

cli-build wakeup 

🎯 Usage

Main Menu

When you start the application, you can choose between:

• CLI Mode — Access Agent, Plan, and Ask modes
• Telegram Mode — Interact with the AI through Telegram
• Exit — Quit the application

---

❓ Ask Mode

Ask questions about your codebase or programming topics.

1. Select Ask Mode
2. Enter your question
3. The AI analyzes the available context and tools
4. Receive an AI-generated answer
5. Optionally save the response if supported

Best for: Quick questions, explanations, debugging, and codebase exploration.

---

🧭 Plan Mode

Create step-by-step implementation plans for larger tasks.

1. Select Plan Mode
2. Enter your goal
3. The AI researches the relevant parts of the workspace
4. Review the generated implementation plan
5. Select the steps you want to execute
6. Confirm before implementation

Example:

Create a REST API for user management

Best for: New features, architectural changes, and complex implementation tasks.

---

🤖 Agent Mode

Give the AI a concrete coding task and let it work through the task using available tools.

1. Select Agent Mode
2. Enter a concrete task
3. The agent analyzes the workspace
4. The agent performs the required operations
5. Review changes before applying them

Example:

Add error handling to the login function

Best for: Coding tasks, refactoring, debugging, and project modifications.

---

💬 Telegram Mode

Interact with the AI through Telegram.

1. Select Telegram Mode
2. The bot starts using your configured Telegram credentials
3. Open your Telegram bot
4. Send messages normally
5. Press Ctrl+C in the terminal to stop the bot

Requires:

TELEGRAM_BOT_TOKEN=your_telegram_bot_token_here
TELEGRAM_OWNER_ID=your_telegram_user_id_here

---

⚙️ Configuration

Model Selection

The model can be configured through CLI configuration or environment variables:

cli-build config set model "nvidia/nemotron-3-super-120b-a12b"

Alternative model:

If nvidia/nemotron-3-super-120b-a12b is unavailable or fails, configure:

cli-build config set model "nvidia/nemotron-3-ultra-550b-a55b"

For .env configuration:

NVIDIA_DEFAULT_MODEL=nvidia/nemotron-3-super-120b-a12b

If needed, replace it with:

NVIDIA_DEFAULT_MODEL=nvidia/nemotron-3-ultra-550b-a55b

Variable	Purpose
NVIDIA_DEFAULT_MODEL	Model used by default for Ask and Agent modes
FIRECRAWL_API_KEY	Enables optional web search
TELEGRAM_BOT_TOKEN	Enables Telegram integration
TELEGRAM_OWNER_ID	Restricts Telegram access to the configured user

You can replace the NVIDIA model IDs with other compatible models supported by your configuration.

💳 Using Paid Models

If you want to use paid AI models or a different AI provider, you can change the AI provider and model configuration in:

ai/ai.config.ts

🛡️ Safety Features

• Staging Area — Changes can be staged before applying them
• Approval Flow — Changes require user approval where configured
• Limited Permissions — Read-only tonols are available where appropriate
• Step Limits — Limits agent iterations to help prevent runaway execution
• Tool Restrictions — Different modes expose different tool capabilities

📝 Project Structure

cli/
├── index.ts                 # Main entry point
├── package.json             # Dependencies and scripts
├── .env.local               # Environment variable template
├── tui/                     # Terminal UI components
│   ├── wakeup.ts            # Banner and mode selection
│   └── terminal-md.ts       # Terminal Markdown rendering
├── modes/                   # Application modes
│   ├── agent/               # Agent mode
│   ├── plan/                # Plan mode
│   ├── ask/                 # Ask mode
│   ├── telegram/            # Telegram mode
│   ├── cli.ts
├── ai/                      # AI provider and model configuration

🔍 Troubleshooting

API key not loaded

The CLI supports both CLI configuration and .env configuration.

For .env, check that:

1. .env exists in the project root.
2. NVIDIA_API_KEY is set correctly.
3. The API key is valid.
4. You restarted the CLI after changing environment variables.

You can safely check whether Bun sees the key without printing it:

bun -e "console.log(Boolean(process.env.NVIDIA_API_KEY))"

It should print:

true

Telegram not working

Check:

• Your Telegram bot token is correct.
• TELEGRAM_OWNER_ID contains the correct numeric Telegram user ID.
• You have started a conversation with the bot.
• The bot is running in Telegram mode.

Slow responses

Response time depends on the selected model, request size, tool usage, reasoning configuration, and current API/server conditions.

For example, different NVIDIA models can have significantly different response times.

If a model consistently responds slowly, try another compatible NVIDIA model.

Missing web search

Set through CLI:

cli-build config set firecrawl-api-key "your_firecrawl_api_key_here"

Or use .env:

FIRECRAWL_API_KEY=your_firecrawl_api_key_here

Then restart the CLI.

✅ Verify the Installation

Run:

cli-build wakeup

Then:

1. Select CLI Mode
2. Select Ask Mode
3. Ask:

What is 2+2?

If the AI responds successfully, your NVIDIA configuration is working.

You can then test Plan Mode:

Explain how to create a basic HTTP server

Finally, test Agent Mode with a small, safe task.

🎉 Getting Started

For your first run:

bun install
bun link
cli-build wakeup

Then try:

1. Ask Mode — Ask a simple programming question
2. Plan Mode — Create an implementation plan
3. Agent Mode — Give the agent a small coding task
4. Telegram Mode — Connect your Telegram bot if configured

---

Built with Bun, TypeScript, and NVIDIA AI.