# Local Multimodal AI Chat

A self-hostable Streamlit chat application that talks to local Ollama models (or optionally OpenAI), with document RAG over PDFs, image chat and Whisper-based voice input.

Local Multimodal AI Chat is a self-hosted chat application built around a single Streamlit interface for text, images, PDFs and voice, with a focus on data privacy and modularity. It runs open models locally through Ollama, with an OpenAI API fallback that can be switched at runtime, and each capability lives in its own handler module.

PDF chat uses retrieval-augmented generation backed by a Chroma vector database, with Ollama's nomic-embed-text for embeddings and configurable chunk size, overlap and retrieved-chunk count. Images are sent to vision-capable models such as LLaVA, and spoken input is transcribed with Whisper. Chat sessions, messages and settings persist in SQLite, and the whole stack is packaged for Docker Compose.

Website: https://github.com/Leon-Sander/Local-Multimodal-AI-Chat