Embedding Service

A local text embedding service using HuggingFace's Text Embeddings Inference server.

Prerequisites

• Docker
• jq (for the CLI tool)
• curl

Quick Start

# Start the service and wait until ready
./setup.sh

# Generate an embedding
echo "Hello world" | ./embed.sh

Configuration

Set environment variables before starting the service:

Variable	Default	Description
EMBEDDING_MODEL_ID	unsloth/embeddinggemma-300m	HuggingFace model to use
EMBEDDING_DIMENSION	768	Embedding vector dimension

Example with a different model:

EMBEDDING_MODEL_ID=BAAI/bge-small-en-v1.5 ./setup.sh

Usage

CLI

# Single string
echo "text to embed" | ./embed.sh

# From a file
cat document.txt | ./embed.sh

API

The service exposes an OpenAI-compatible API on port 8080:

curl http://localhost:8080/v1/embeddings \
  -H 'Content-Type: application/json' \
  -d '{"model":"unsloth/embeddinggemma-300m","input":"text to embed"}'

Stopping

docker compose down