LocalAgent – Terminal AI Assistant (Prototype)

A lightweight terminal-based AI agent prototype powered by Ollama, created to explore how local language models can assist with chat, code analysis, and basic file operations.

---

Overview

LocalAgent is a minimal, experimental implementation designed to understand:

• How terminal AI agents operate
• How local LLMs (via Ollama) respond to structured prompts
• How AI-assisted file and code generation works in practice

This project is intentionally scoped as a learning-focused prototype, not a production system.

---

Scope & Intent

• ✅ Built to experiment and learn
• ✅ Demonstrates core agent workflows
• ✅ Useful for general chat purpose
• ❌ Not production-ready
• ❌ No guarantees of correctness or safety

This prototype is meant for internal evaluation, learning, and discussion.

---

Features

• Multi-Model Support
  Compatible with any locally available Ollama model

• Chat Mode
  General-purpose AI chat and explanations

• Build Mode
  Generates files and folder structures
  (general-purpose scaffolding only)

• Analyze Mode
  Basic codebase inspection and statistics

• Safe File Operations
  Requires confirmation before creating files

• Rich Terminal UI
  Uses rich for readable and interactive CLI output

---

Important Notes & Safety

• Build mode may be slow, depending on hardware
• Local models can be outdated or inaccurate
• Generated files may be incomplete or incorrect
• Outputs should always be reviewed manually
• Not suitable for production or automated deployment

Chat mode is generally reliable for exploration and discussion,
but build outputs should be treated as assistive suggestions only.

---

Quick Start

Prerequisites

• Ollama installed and running
• Python 3.8+

Installation

git clone https://github.com/Adii0906/Local-Agent-CLI.git
cd localagent
pip install -r requirements.txt
python agent.py

Commands

Command Description

/chat	Interact with the AI
/build	Generate files/folders
/analyze	Analyze a codebase
/model	Switch Ollama model
/help	Show help
/exit	Exit the agent

Recommended Models

qwen2.5-coder:1.5b – Lightweight, good for coding
deepseek-coder:1.3b – Strong code-focused model
mistral:7b – Fast general-purpose model

---

Acknowledgements

Built using Python and Rich for an interactive and user-friendly experience.

Special thanks to Ollama for providing powerful free local AI models, making local experimentation and learning possible without relying on cloud-based APIs.