Tichy

Detailed Introduction

Tichy is a self-contained, privacy-focused retrieval-augmented generation (RAG) system implemented in Go. It provides a complete local pipeline for vectorized retrieval and inference: PostgreSQL with pgvector for storage, local embedding and inference via llama.cpp, and document ingestion and indexing tools. By keeping models and data on-premises, Tichy is suitable for use cases that require data privacy and operational control.

Main Features

• Local-first deployment: run all data and models locally without relying on external LLM providers.
• Full RAG pipeline: document chunking, embeddings, vector indexing, retrieval, and inference via llama.cpp.
• Containerized services (Docker Compose) for engineering-friendly deployment and scaling.
• CLI tools for ingestion, interactive chat, tests generation, and evaluation.

Use Cases

• Privacy-sensitive QA systems and chat assistants that keep user data locally while enabling contextual retrieval.
• Enterprise or research knowledge base construction and offline RAG evaluation.
• Edge or network-constrained environments using local GGUF models for embeddings and inference.

Technical Features

• Semantic embedding-based retrieval backed by pgvector for efficient vector queries.
• Engineering-oriented Go implementation with HTTP-compatible service interfaces.
• Optional GPU acceleration (llama.cpp + CUDA) or CPU mode for portability and performance trade-offs.