I found it frustrating that there wasn't a simple, easy to use program for studying program in an interactive or kinesthetic way. So, while I worked on other projects, I decided to slop together a program and leverage local LLMs to build TactiTerm.

A self-sufficient, offline capable IDE, with an LLM powered tutor to help guide you through mental roadblocks, and build up confidence when it comes to programming in various languages.

Overview

TactiTerm is an interactive software engineering tutor designed with a "guidance over answers" philosophy. Rather than providing copy-paste solutions, TactiTerm utilizes an LLM Harness to ask probing questions, provide hints, and guide users toward mastering concepts across various languages.

TactiTerm features both a rich Terminal User Interface (TUI) for command-line use, and a modern Web UI featuring the Monaco Editor.

Features

AI Mentor: Provides step-by-step guidance, code review, and conceptual hints without spoiling direct answers.
Dual Interfaces:
- TUI: Built with Python and Textual for lightweight, keyboard-driven terminal workflows.
- Web UI: Built with React, Vite, TypeScript, Tailwind CSS, and Monaco Editor.
10 Supported Languages: Python, C#, C++, Java, JavaScript, TypeScript, Rust, Lua, HTML, and Go.
Real-time Linting & Execution: Tree-sitter powered syntax validation and safe local code execution.
Interactive Syntax Handbook: Comprehensive reference catalog with on-demand AI code examples for functions and core language topics.
AI Challenge Generator (GenTUI): Create and save custom Markdown programming challenges tailored to specific subjects and difficulties.
Configurable & Network Ready: Easily configure LLM endpoints, ports, and enable "public": true to host the Web UI across your local network (0.0.0.0).

Quick Start

Prerequisites

Devbox

Currently we only officially support running TactiTerm through devbox. This is to simplify the deployment process across devices. If you dare, and you have the prequisites installed; you should be able to run TactiTerm without it.

Installation

Clone the repository:

git clone https://github.com/your-username/TactiTerm.git
cd TactiTerm

Install dependencies:
```
devbox run install
```
(or make install inside devbox shell)

Usage

TactiTerm is executed via the use of Devbox run commands.

Terminal Interface (TUI)

Launch the interactive terminal application:

devbox run tui

(or make tui inside devbox shell)

Additional TUI modes:

devbox run gentui - Launch the standalone GenTUI AI Challenge Generator.
devbox run -- make tui-debug - Launch TUI in debug logging mode.

Web Interface (Web UI)

Launch the FastAPI backend server and Vite frontend dev server:

devbox run web

(or make web inside devbox shell)

Then open your browser to http://localhost:5173.

Configuration

When initialized, TactiTerm automatically generates a config.json file in the root directory if one does not exist.

`config.json` Example

{
  "llm": {
    "base_url": "http://localhost:8080/v1",
    "model": "local-model",
    "api_key": "not-needed",
    "temperature": 0.7,
    "max_tokens": 512,
    "timeout_seconds": 60.0
  },
  "web": {
    "host": "127.0.0.1",
    "port": 8000,
    "public": false
  }
}

Key Settings

LLM Settings: Point base_url to any OpenAI-compatible server (LocalAI, vLLM, Ollama, LM Studio, or OpenAI API).
Web UI Public Access: Set "public": true under "web" (or set environment variable TACTTERM_WEB_PUBLIC=true) to bind the server to 0.0.0.0, allowing other devices on your local network to connect.

Environment Variable Overrides

TACTTERM_LLM_BASE_URL - Override LLM API URL.
TACTTERM_LLM_MODEL - Override LLM model name.
TACTTERM_WEB_HOST - Override web server bind address.
TACTTERM_WEB_PORT - Override web server port (default: 8000).
TACTTERM_WEB_PUBLIC - Set to true to enable network binding (0.0.0.0).

Architecture

Backend (src/api/main.py): FastAPI application serving REST endpoints for challenges, code linting, code execution, mentoring guidance, and handbook references.
Core Engine (src/core/):
- config.py: Application config manager with auto-generation and env overrides.
- executor.py: Subprocess execution engine for multi-language code snippets.
- linter.py: Tree-sitter syntax checker.
- mentor.py: Async Socratic mentor prompt & LLM service.
- handbook.py: Reference catalog and example generator.
- generator.py: AI challenge generation service.
TUI Client (src/tui/): Textual terminal application.
Web Client (frontend/): React + TypeScript SPA powered by Vite, Tailwind CSS, and Monaco Editor.

README.md

TactiTerm