Practical overview
When it fits
Use it when you need question answering over a folder of supported local documents and want answers linked to relevant source files.
Who it is for
- People who want to ask natural-language questions across their own document folders.
- Developers who want a local HTTP service or Streamlit interface for document question answering.
Documented uses
- Index supported files in a folder and its subfolders.
- Search indexed documents semantically and get generated answers with references to relevant documents.
- Expose document questions through the provided local HTTP endpoints or Streamlit interface.
Requirements & limits
The repository is public and lists GPL-3.0 as its license. Running the application requires local setup; the README describes Hugging Face access and an NVIDIA NIM API key for the NVIDIA inference route.
- The indexed file types are PDF, TXT, DOCX and PPTX.
- The README describes NVIDIA API-backed 70B inference and a local 8B model option.
- The documented interfaces are a local HTTP service and a Streamlit UI.
- The README gives a 32 GB computer as the development setup and cautions that model execution needs a sufficiently powerful machine.
Check current requirements in the official documentation. Source-based guidance does not mean OpenViST has run or independently tested this project.
Index a folder of supported documents
Create a local Qdrant index from a folder and its subfolders.
Before you start
- A local copy of the repository and its installed Python requirements.
- An environment_var.py file containing a Hugging Face API key.
- A folder containing PDF, TXT, DOCX or PPTX files.
Steps
Install the project requirements with pip install -r requirements.txt.
Create environment_var.py and add the Hugging Face token as described in the README.
Run python index.py path/to/folder, replacing the path with the folder to index.
Expected outcome
The application creates a local Qdrant index for supported files in the specified folder and its subfolders.
Guide limitations
- This covers indexing only; the README describes running the search service and interface as separate next steps.
- The README identifies supported extensions as PDF, TXT, DOCX and PPTX.
Start the local question-answering interface
Start the Streamlit interface for asking questions about indexed local files.
Before you start
- An indexed folder.
- Installed project requirements and the configured Hugging Face token.
- Access to the Llama 3 model; the README describes a local 8B option and an NVIDIA API-backed 70B option.
Steps
For the NVIDIA route, add the NVIDIA NIM API key to environment_var.py.
Start the service with python uvicorn_start.py.
Start the interface with streamlit run user_interface.py.
Expected outcome
A Streamlit interface is available for asking questions based on the indexed files.
Guide limitations
- The README says model loading, including downloads from Hugging Face, may take several minutes, especially the first time.
- The README describes the local 8B option as requiring a sufficiently powerful machine and gives a 32 GB computer as its development setup.
Sources & review
Source record details
- Source availability
- Active
- Sources checked
- 17 October 2026
- Original website title
- nikolamilosevic86/local-genAI-search · Official homepage title · retained in its source language