Agentic CLI Starter

Overview

Starter code for a CLI-based AI agentic chatbot with tool-calling capabilities. Vibe coded this to understand apps like Claude Code better.

Ways to extend / configure:

• Add more tools
• Add more complex history management logic
• Add more LLM models
• Better agentic reasoning

High level inner-workings

• index.ts starts a Read-Event-Print Loop
  • repl.ts responsible for coordinating user input, agents, and UI
• REPL calls our agent
• Agent calls LLM with chat history + tools
• On a tool call, our agent calls the respective tools and appends results to chat history
• Repeat agent call until agent does not need a tool call
• Once agent responds without tool call, REPL updates UI to inform user the results
• The loop starts again

Quick Start

Add your OpenAI API Key in a .env to get things working.

# Install dependencies
npm install

# Start the interactive chat
npm run start

Features

• Interactive Chat: Continuous conversation with the AI agent
• Slash Commands: Special commands for controlling the CLI
• Tool Integration: Agent can use tools (web search) to answer queries
• Conversation History: Maintains context across multiple turns
• Visual Feedback: "Thinking..." indicator while agent processes
• Error Handling: Tool or LLM failures are accounted for.
• File logging: Logs are saved to file per chatting session using a custom logger.

Usage Example

Chat started. Type /help for commands or /exit to quit.

You: What is the weather like today?
Agent: thinking...
Agent: [Agent uses web search tool and responds with weather info]

You: /info

Session Info:
  Messages in history: 3
  Tools available: web_search

You: /exit
Goodbye!

Architecture

See AGENT.md for detailed architecture documentation.