RepoChat in 200k - No RAG!

⚡Chat with GitHub Repo Using 200k context window of Claude instead of RAG!⚡

Why RepoChat - No RAG?

Take the advantage of Claude 200k! Put all examples and codes to the contexts!

We need copilot rather than agent sometimes!

1. Not just Chat, But Write Codes with the Lastest Documents

Have troubles memorizing all the apis in llama-index or langchain?

No worries, just include the components examples and the documents from the repo and let Claude Opus - the strongest model and long context window of 200k to write your agent for you!

2. No copy paste, Just Select

Download/Clone your Repo from Github then just select the files you'd like, I got you covered on constructing the prompt.

3. No RAG, You decide

I've seen many Chat with Repo projects, they all have the painpoints:

Which files do this query need?

They use embedding search in Code database but most of the time I already knew what documents I'm refering... So make your own choices each time when you are coding.

• Coding Frontend? Just select components and examples.

• Coding Agents? Just select Jupyter Notebook of langgraph.

• Coding RAG? Just select Jupyter Notebook of llamaindex.

4. More Use Cases!

Use llamaindex doc

select llamaindex example of pipeline to write RAG graph.

Use Shadcn doc to write frontend

select examples and components definition.

Suggestions!

1. You can use Haiku for most of the case.
2. Change models based on tasks.
3. Change files based on tasks.
4. Clone Repos You like!
5. Usually I will include README of repo to help Claude Understand better
6. USE COUNT TOKENS on the sidebar to see how many tokens you will send!!!

Features

1. Repository Download: Users can provide a GitHub repository URL, and the application will automatically download and analyze the repository.
2. File and Folder Selection: Users can select specific files or folders from the repository to include in the LLM's input.
3. Language Filtering: Users can filter the files by programming language to focus the LLM's understanding on specific parts of the codebase.
4. Token Limit: Users can set a token limit to control the amount of information sent to the LLM, which can be useful for performance or cost considerations.
5. Chat Interface: Users can interact with the LLM through a chat-style interface, allowing them to ask questions or request code generation based on the repository contents.
6. Streaming Output: The LLM's responses are displayed in a streaming fashion, providing a more engaging and real-time user experience.

Currently I only supported Openrouter. Planing to add more and refactor someday.

Get Started

1. Environment Settings: Run pip install -r requirements.txt to set up environment.

2. Create a .env file: Create a .env file in the root directory of the project and add your OpenRouter API key (Recommended):

OPENROUTER_API_KEY=your_openrouter_api_key_here

I recommend OpenRouter because it has all models!

If you want to use OpenAI GPT models, add your openai api key as well.

OPENAI_API_KEY=your_openai_api_key_here

3. Run the application: Run the app.py script using Streamlit:

streamlit run app.py

4. Use the application: Follow the instructions in the application to download a GitHub repository, select files and folders, and chat with the LLM.

If you encounter some issues with repo, you can always delete the repo dir in ./repos dir and download it again.

Configuration

The application's behavior can be customized through the following configuration options:

• Model: The specific LLM model to use (e.g., "anthropic/claude-3-haiku", "anthropic/claude-3-opus").
• Temperature: The temperature parameter that controls the "creativity" of the LLM's responses.
• System Prompt: The initial prompt given to the LLM to set the desired behavior.

These settings can be adjusted in the sidebar of the Streamlit application.

Contributing

If you'd like to contribute to the RepoChat-200k project, please feel free to submit issues or pull requests on the GitHub repository.

License

This project is licensed under the MIT License.