datashaman/code-butler

OpenAI assistants that help you write code.

code-butler

OpenAI assistants that help you write code.

The app is an interface to define assistants and projects. The AI assistants have access to file contents and can make changes to them (using OpenAI Function Calling).

code-butler...

• is a proof of concept
• is a work in progress
• makes destructive changes to your project files
• is not responsible for any data loss or corruption

Use with caution and source control.

You do not have to point the assistant to your whole project directory. You can point it to a subdirectory, focusing on just components (for example). It is up to you to decide how much access you want to give the assistant. Less is better, less room to make mistakes.

table of contents

• roadmap
• installation
• configuration
• database migration
• ssl certificate
• development
• prompts
• architecture

roadmap

• create project
• create assistant
• chat with assistant
• undo changes - added simple git commit and revert
• project facts - AI managed project facts, used to prompt the assistant
• run steps - UI for the run steps
• direct voice control - currently it transcribes voice to text, which the user sends. It should send the text directly to the assistant on transcription.
• continuous voice control - it now posts the message automatically, but it would be neat to have it always-on with idle timeout
• fix reactivity of run steps
• @mention - assistant mentions in chat
• uploads - upload files to the assistant
• project file tree - to drag filenames to the chat for the AI to focus on
• ignore files - list of files/folders which the assistant should ignore and will not be returned in the file tree (for example node_modules or vendor) or file tools
• playwright - use playwright to browse the project, test changes, take screenshots, etc.

installation

Install the GitHub CLI:

brew install gh

Clone the repository:

gh repo clone datashaman/code-butler
cd code-butler

Install dependencies:

yarn

configuration

Copy .env.example file to .env:

cp .env.example .env

Edit the .env file and set OPENAI_API_KEY to your OpenAI API key.

database migration

This creates and migrates a SQLite database database/db.sqlite:

 yarn db:migrate

ssl certificate

The app requires an SSL certificate for the voice input to work.

First install mkcert:

brew install mkcert

Add the certificate authority to the system trust store:

mkcert -install

Then generate the certificate:

cd certs
mkcert localhost

development

Start the development server:

yarn dev

This will start the server at https://localhost:3000.

There should be no warning about the certificate. If there is, ensure you added the certificate authority to the system trust store as decribed above.

mkcert -install

prompts

The assistant instructions are important. They form the basis of the assistant's knowledge and capabilities, and are used to generate the assistant's responses. All tools should be selected when creating, and updating the assistant.

The project description is important. It is sent as additional instructions to the assistant when it is called within a project.

Use this to provide context and constraints for the project. Describe the project requirements, the programming language, the libraries and frameworks used, and any other relevant information that will help the assistant understand the project.

The model dropdown on the chat page is for overriding the default model used by the assistant assigned to the project. Set it to assistant to use the default model.

Ensure the path field is set to the correct path to the project directory. Sadly we cannot use a directory picker in the browser, so you will have to type the path manually or copy it from the console.

architecture

The project is a Nuxt.js app with a SQLite database.

Tailwind CSS and DaisyUI are used for styling.