mlaude-engine

Core RAG engine for Mlaude — a fully local, private personal AI agent that answers questions about your goals, decisions, reflections, and technical learnings.

Stack

• LLM: Llama 3.1 8B (Q4_K_M) via Ollama
• Embeddings: paraphrase-multilingual-MiniLM-L12-v2
• Vector DB: ChromaDB (file-based)
• RAG Framework: LlamaIndex
• Speech-to-Text: Whisper small
• CLI: Typer + Rich

Setup

# Install dependencies
uv sync

# Pull Ollama model
./scripts/setup_ollama.sh

# Download Whisper model
./scripts/setup_whisper.sh

Usage

# Interactive chat
mlaude chat

# Ingest documents
mlaude ingest <path>

# Check status
mlaude status

Architecture

All data stays local. The engine reads from a separate knowledge repo (configured via ~/.mlaude/config.yaml) and stores vector embeddings in ~/.mlaude/chroma_db/.