Below is a step-by-step walkthrough of the app:

Step 1	Step 2

Step 3	Step 4

Step 5	Step 6

Follow these simple steps to get started with Offline Doc Chat:

1. Set Your Ollama Endpoint and Model: Navigate to the Settings panel and configure the connection to your locally hosted Ollama instance.

2. Upload Your Documents: Drag and drop your files into the upload area. The system will automatically process and embed them in the background.

3. Start Asking Questions: Once processing is complete, you can ask natural language questions about your documents — all without internet access!

You can fine-tune the entire RAG (Retriever-Augmented Generation) pipeline via Settings > Show Advanced Options. This gives you full control over model behavior and retrieval strategy.

Local RAG is powered by llama-index and utilizes its SimpleDirectoryReader() to process a wide variety of document types (e.g. .pdf, .md, .ipynb). Here's what happens under the hood:

• Each file is split into smaller logical units. For example, a multi-page PDF is separated into one document per page.
• These documents are then chunked based on the configured chunk_size, with optional chunk_overlap to preserve context.
• The resulting chunks are embedded using the selected model and stored for fast retrieval.

You can tweak these settings to balance between embedding precision, retrieval speed, and system performance.

💡 Tip: Experimenting with different chunk_size and chunk_overlap values can significantly affect answer quality. A smaller chunk size may be better for detailed queries, while larger chunks can improve speed.