aicodec 1.4.0

A lightweight communication layer for developers to interact with LLMs.

AI Codec

AI Codec is a lightweight, CLI-first tool designed to streamline the interaction between a software developer and a Large Language Model (LLM). It provides a structured, reviewable, and reversible workflow for applying LLM-generated code changes to your project.

It’s designed for the developer who:

• Prefers interacting directly with an LLM's web chat window.
• Wants to avoid the complexity and cost of managing API keys.
• Is not using a fully-integrated solution like Aider and wants a structured way to apply code.

The Problem

LLMs are incredibly powerful for generating code, but integrating their suggestions into a project is often a messy, manual process. Developers typically face:

• Unstructured Output: LLMs produce large, unstructured blocks of code that are difficult to parse.
• Tedious Manual Work: Manually copying and pasting changes across multiple files is slow and highly prone to error.
• Difficult Reviews: It's hard to see a clear "diff" of the proposed changes before applying them, losing the safety of a code review process.
• No Easy Undo: If an applied change introduces a bug, there is no simple, one-step process to revert the file system to its previous state.

Why Use AI Codec?

AI Codec solves these problems by treating LLM-generated changes as a formal, reviewable patch, much like a pull request.

• Structured Interaction: It enforces a simple JSON schema, turning the LLM's raw output into a structured set of changes.
• Safe Review Process: The aicodec apply command launches a web UI that provides a git-like diffing experience, so you can see exactly what will change before any files are touched.
• Developer in Control: You have the final say. Selectively apply or reject any change, or even edit the LLM's suggestions live in the diff viewer.
• Atomic & Reversible Changes: The apply and revert commands make applying LLM suggestions a safe, atomic transaction that you can undo with a single command.

Features

• Interactive Project Setup: Quickly initialize your project with aicodec init.
• Flexible File Aggregation: Gather all relevant project files into a single JSON context for the LLM, with powerful inclusion/exclusion rules.
• Web-Based Visual Diff: Review proposed changes in a clean, web-based diff viewer before they are applied to your file system.
• Selective Application: You have full control to select which files to modify, create, or delete from the LLM's proposal.
• One-Click Revert: Instantly revert the last set of applied changes with the aicodec revert command.
• Clipboard Integration: Pipe your LLM's response directly from your clipboard into the review process.
• Built-in Schema Access: Easily access the required JSON schema with the aicodec schema command.

---

Installation

Aicodec is available on PyPi.

pip install aicodec 

This will make the aicodec command available in your terminal.

Workflow and Usage

The aicodec workflow is designed to be simple and integrate cleanly with your existing development practices, including version control like Git.

Step 1: Initialization

First, initialize aicodec in your project's root directory. You only need to do this once.

aicodec init

This command will guide you through an interactive setup to create a .aicodec/config.json file.

Step 2: Aggregating Context

Next, gather the code you want the LLM to work on.

aicodec aggregate

This command scans your project based on your configuration and creates a context.json file. This file contains the content of all relevant files, which you can now provide to your LLM.

Step 3: Generating the Prompt

Run the prompt command to generate a ready-to-use prompt file that includes your aggregated context:

aicodec prompt

This will create a prompt.txt file in your .aicodec directory (configurable). You can customize the task with --task "your description" or copy directly to clipboard with --clipboard.

Step 4: Generating Changes with an LLM

Copy the contents of prompt.txt (if not already in your clipboard) and paste it into your LLM of choice. Ask it to perform refactoring, add features, or fix bugs.

Crucially, you must instruct the LLM to format its response as a JSON object that adheres to the tool's schema.

You can then provide this schema to the LLM along with your project context and prompt.

Step 5: Preparing to Apply Changes

Once you have the JSON output from the LLM, copy it to your clipboard.

Then, run the prepare command:

aicodec prepare --from-clipboard

This validates the JSON from your clipboard and saves it to .aicodec/changes.json, getting it ready for review. You can also configure it as default flag in your config by adding under "prepare" the "from-clipboar": true option.

Step 6: Reviewing and Applying Changes

This is the most important step. Run the apply command to launch the web-based review UI:

aicodec apply

Your browser will open a local web page showing a diff of all proposed changes. Here you can:

• Select or deselect individual changes.
• View a color-coded diff for each file.
• Edit the proposed changes directly in the UI.

Once you are satisfied, click "Apply Selected Changes". The tool will modify your local files and create a .aicodec/revert.json file as a safety net.

Step 7: Reverting Changes (The "Oops" Button)

If you are unhappy with the result of an apply operation, you can easily undo it.

aicodec revert

This command opens the same review UI, but this time it shows the changes required to restore your files to their state before the last apply operation.

Additional commands

Get schema

To get the required schema, which the llm needs to follow, run the following command in your terminal:

aicodec schema

You can directly pipe this output to your clip by e.g. using aicodec schema | pbcopy on macOS, | clip on Windows, or | xclip on Linux.