FeatureTranscribeAI is an innovative tool designed to streamline the development process by automating the identification and generation of code based on new feature requests. Utilizing the power of machine learning and the OpenAI API, FeatureTranscribeAI does the following:
- Generates embeddings from your codebase
- Analyzes user input for feature requests
- Identifies relevant existing code
- Crafts prompts for generative models to automatically generate code fulfilling the feature request.
Find relevant files with a prompt
find.files.mov
Write code WITH your code as context
write.tests-new.mov
Continue the converation and add context
add.context.mov
- Automated Embeddings Generation: Generate embeddings from your existing codebase to understand and analyze your code at a deeper level.
- Feature Request Processing: Accepts user inputs for new features and uses machine learning to understand the request.
- Relevant Code Identification: Finds relevant code in your codebase that aligns with the feature request, using generated embeddings and machine learning algorithms.
- Code Generation for New Features: Crafts prompts based on identified relevant code and the feature request for generative models, facilitating the automatic generation of code snippets to implement new features.
curl -s https://raw.githubusercontent.com/miltonian/feature-transcribe-ai/main/install.sh | bash- feature_to_code.py: Analyzes embeddings to identify code snippets relevant to a new feature request. It handles loading embeddings, code content, and new feature descriptions.
- openai_api.py: Interfaces with the OpenAI API for generating embeddings from text inputs, crucial for analyzing both existing code and feature requests.
- prepare_code.py: Prepares your codebase for analysis by parsing and generating embeddings, integrating functionalities from
openai_apiand other parsing utilities. - code_parser.py: Specializes in parsing Swift/Typescript/Javascript/Python files, enabling detailed analysis and embeddings generation for Swift/Typescript/Javascript codebases.
- utils.py: Offers utility functions for file and directory management, supporting the overall workflow of FeatureTranscribeAI.
- Typescript
- Javascript
- More coming soon...
- Python 3.x
- OpenAI API Key
Using FeatureTranscribeAI involves just a couple of straightforward steps. Before starting, ensure your environment is set up as described in the Setup section.
-
Export your OpenAI Api Key
export OPENAI_API_KEY=<your-key>
-
Sync your project: This will prompt you to enter your project's directory. This script will analyze your codebase and generate embeddings. This process respects your
.gitignoresettings, automatically excluding files and directories you've opted not to track with Git. This ensures a focus on the meaningful parts of your codebase for analysis.sh sync.sh
-
Starting coding with AI: Once your codebase is prepared, run
sh run.shwith a description of the new feature you want to integrate. The script uses the generated embeddings to find code snippets that are most relevant to your feature request.sh run.sh
When prompted, you can interact with the tool in various ways, depending on the context of your request and the information you provide:
- First Message to the AI: If it's your first interaction in a session, the AI needs context to understand where to locate or suggest relevant code. You can either:
- Provide Detailed Context: Include both the filename or path and a clearly identifiable code snippet (e.g.,
describe("Some test case"),router.put("/some/endpoint"), orexport const someVariableOrFunction). This helps in pinpointing the exact location for implementing or modifying code. - General Inquiry: If you don’t provide specific details, the AI will make an educated guess based on its understanding of the codebase and indicate where the relevant code might likely exist.
- Provide Detailed Context: Include both the filename or path and a clearly identifiable code snippet (e.g.,
- Adding Code Context: If you are continuing a conversation and wish to add specific code context to refine or expand upon a previous inquiry, ensure to provide both the filename or path and a specific code snippet. This will help the AI to accurately address your new context or modify its previous responses.
- General Discussion: If specific details are not provided after the initial context has been set, you can freely discuss other aspects or follow up with general inquiries. The AI will base its responses on the accumulated context of the conversation, using any prior detailed information you've provided to inform its responses.
- You are not restricted to only providing detailed context; you may ask questions or make requests in any form. However, the accuracy and relevance of the AI's response can be significantly enhanced by how specific the provided information is, especially in your initial interaction or when adding new context.
FeatureTranscribeAI automatically skips over files and directories specified in your project's .gitignore file during the embeddings generation and relevant code identification process. This ensures that temporary files, dependencies, or any other non-relevant code specified in .gitignore are not included in the analysis, keeping the focus on the meaningful parts of your codebase.
After the first execution of sh sync.sh, it will only generate updated embeddings for code that has been updated since your last sync
This feature helps maintain the integrity and relevance of the embeddings and code suggestions, providing more accurate and useful results based on the content you actively maintain in your repository.
Contributions to FeatureTranscribeAI are highly encouraged. Whether it's by enhancing functionality, adding new features, or fixing bugs, your input is valuable. Please feel free to fork the repository, make your changes, and submit a pull request.
FeatureTranscribeAI is released under the MIT License. For more details, see the LICENSE file in the repository.