aiCoder

Harness AI to Craft Your JavaScript Libraries with Ease!

Tired of manually copying and pasting code snippets from ChatGPT into your editor? aiCoder is your solution, streamlining the coding process with an AI-powered assistant.

With aiCoder as your coding co-pilot, you can simplify development, refine your code, and bring ideas to life through natural language interactions. It intelligently merges AI-suggested changes into your existing code—without disrupting what you’ve already built.

The magic lies in aiCoder’s advanced merging logic, which keeps your original code intact while integrating new snippets precisely and reliably.

Slide Deck discussing the AST approach to merging snippets

Code manipulator source file

Purpose and Benefits

aiCoder is designed to enhance your coding experience by leveraging the power of AI. It helps you:

• Save time by automating repetitive coding tasks.
• Improve code quality with intelligent suggestions.
• Focus on higher-level design and problem-solving.
• Seamlessly integrate new features and improvements into your existing codebase.

UNSTABLE

Note: This project is under rapid development. Settings and prompt files may need to be recreated as updates are released, since more settings and file renames are on the horizon.

Standard Workflow

1. Run aiCoder in your project’s directory.
2. Visit http://localhost:3000 to access the web UI.
3. Use New Plan Chat to discuss and define your project requirements, then save the plan. The saved plan guides future implementations.
4. Open the Files tab to select the target file you want to modify.
5. Start a New Chat with the selected file.
6. Request code changes using natural language prompts.
7. Review the suggested snippets and click 🤖✎⚡ to automatically merge them into the file.

Installation

Installation instructions provided for linux.

From NPM

npm i -g aicodertool

From Source

1. Clone this repository and navigate to the cloned directory.
2. Run:

   cd aiCoder
   npm install && npm install -g .
   

After installation, you’re all set! Just run the aiCoder or the much short a command in any project directory.

Installing Ollama allows you to use local LLMs. On linux you can run aiCoder -setup to automatically install ollama and pull the default model or manually install it with the following commands:

# Install Ollama 
curl -fsSL https://ollama.com/install.sh | sh

# Pull the preferred model
ollama pull granite3-dense

You can find and install additional models to test with https://ollama.com/search

Usage

Launch aiCoder from within your project folder:

cd myProjectFolder
aicoder

Then open http://localhost:3000 in your browser to begin.

To run on a different port use aicoder -p 3001

Chat Interface

aiCoder provides two different specialized chat modes:

New Chat

Use New Chat to interactively modify a selected file. Before starting a conversation, be sure to choose a file from the Files tab. The LLM uses a system prompt specifying the snippet format. Correctly formatted snippets can be merged automatically, updating existing functions or class methods without disturbing the rest of the code.

New Plan Chat

Use New Plan Chat to refine the project plan document. This plan, combined with the default system prompts, guides the LLM during regular chats.
Begin by describing your project goals. You can ask the LLM to produce a detailed outline of all the necessary classes and methods, then save this plan.

Tools

Methods List

Displays all detected methods.

• Green methods: Implemented methods with code.
• Red methods: Stubs that need implementation.

Clicking on a method passes it into a new chat, allowing you to modify or implement it.

Stubs List

Shows only stub methods, making it easy to identify and implement missing functionality.

Merge and Format

If you’re using ChatGPT or another external LLM, you can copy and paste suggested snippets to the end of the file. They must follow the guidelines in snippet_production_prompt.md for automatic merging.

Prepend Class Structure

Adds a skeleton of all existing classes and methods to the top of the file. You can reorder classes and methods here. After adjusting the structure, run Merge and Format to clean up and integrate the changes into the stubs.

Project Settings

Project-specific settings are stored in ./.aiCoder/. This folder includes backups of modified files stored in ./.aiCoder/backups/.

You can customize system prompts for your project in the ./aiCoder/ folder. Access these settings in your IDE or from the Project Settings tab.

default_plan_prompt

Describes your project, goals, and requirements. Edit it directly, or use New Plan Chat to interactively refine it.

default_system_prompt

Shapes the LLM’s personality and behavior. If you break it, don’t worry—delete it and restart aiCoder to restore defaults. If you find improvements, consider opening an issue with your suggested changes.

snippet_production_prompt

Instructs the LLM on how to format code snippets for seamless merging.
All snippets must conform to the following format:

class exampleClass {
   // ... existing methods
   exampleMethod(){
      //example code
   }
}

LLM Settings

Before using aiCoder, configure your LLM provider and model. After providing an API key, you can view available models.

Choosing an LLM Provider

• OpenAI’s gpt-4o-mini or gpt-4o models work well.
• Anthropics’s claude-3-5-haiku is an option.
• Groq is less tested and has a smaller context window.
• Ollama runs locally but may be slower. Its granite3-dense:8b model works well enough, though sometimes requires additional prompting to get properly formatted snippets.

Backup Functionality

Every time aiCoder modifies a file, it creates a backup in ./.aiCoder/backups/, preserving your original code with timestamped filenames.

✅ Current State of the Project

Implemented:

• Automatic merging of multiple class/function definitions into a single source structure.
• Ability to prompt for modifications to a single JS file.
• Automatic backups before modification.
• Code snippet extraction from LLM responses and a workflow to accept/reject on a per-snippet basis.
• Custom prompt manager for reusable prompts.
• Support for various LLM providers.
• Automatic setup on first use (provider selection, API key, model choice).
• NPM packaging.
• Automatic generation of stub methods and their code.

To Do:

• Support multiple files simultaneously.
• Allow the LLM to suggest adding NPM packages.
• Add support for HTML, CSS, and MD files by parsing them to ASTs and merging changes.
• Preserve comments during merges.

🌌 Looking Ahead

aiCoder is more than a tool—it’s your personal AI development partner. By making coding smarter, faster, and more fluid, aiCoder helps you unlock new creative possibilities.

Embark on your journey with aiCoder today, and experience a whole new world of coding! 🌟

FAQs

1. How do I reset my settings?
   • Delete the ./.aiCoder/ folder and restart aiCoder. This will recreate the default settings and prompts.
2. Can I use aiCoder with other programming languages?
   • Currently, aiCoder is optimized for JavaScript. Support for other languages is planned for future releases.
3. How do I update aiCoder?
   • If installed via npm, run npm update aicodertool -g. If installed from source, pull the latest changes from the repository and reinstall.