Lavabo

Lavabo is an all-in-one Docker container that simplifies local AI development. It provides unified HTTP endpoints to run LLMs, text and image embedding models, vision models, and TTS models, all within a single environment.

By focusing on small, locally-hosted models, Lavabo makes powerful AI broadly accessible. The name, "Lavabo," is Spanish for washbasin, reflecting the "kitchen sink" approach to providing a complete inference solution for compact models.

Lavabo provides out-of-the-box support for a versatile range of compact models:

• Any GGUF LLM: Use any local or Hugging Face-hosted GGUF model for tasks ranging from standard chat to complex structured output.
• Transformers for Embeddings: Easily create text embeddings with built-in helper functions for similarity search.
• Dual Text-to-Speech Engines:
  • Kokoro TTS: A compact model perfect for quick integration with multiple built-in voices.
  • Piper TTS: Offers a large selection of voices and accents and supports training new voices for maximum flexibility.
• CLIP for Image Classification: Categorize images using simple and intuitive natural language prompts.
• Moondream for Advanced Vision: Go further with a powerful Vision-Language Model (VLM) for image captioning, visual Q&A, object detection, and pointing.

Quick start

• Build: docker compose -f server/docker-compose.yml build
• Run: docker compose -f server/docker-compose.yml up

Prebuilt Docker image (GHCR)

You can pull and run the prebuilt image from GitHub Container Registry.

• Pull the latest image from main:
  • docker pull ghcr.io/zetaphor/lavabo:main
• Or pull a pinned release (when available):
  • docker pull ghcr.io/zetaphor/lavabo:vX.Y.Z

Run with NVIDIA GPU (recommended):

docker run --rm -it \
  --gpus all \
  -p 8000:8000 \
  -e GGML_CUDA=1 \
  -e N_GPU_LAYERS=-1 \
  -v /home/zetaphor/LLMs:/models \
  ghcr.io/zetaphor/lavabo:main

Notes:

• The API will be available at http://localhost:8000 (see Swagger UI at /docs).
• Mount your local GGUF models under /models in the container (adjust the host path as needed).
• This image targets CUDA; ensure the NVIDIA Container Toolkit is installed and GPUs are visible to Docker.
• For detailed API endpoints and usage, see server/README.md.