Install text-embeddings-inference with Homebrew

Project links

• Homepage
• Repository
• Upstream docs
• Package manager page

Commands and aliases

• text-embeddings-router

Project history

The official repository and documentation describe TEI as a toolkit for deploying and serving open source text embeddings and sequence classification models. Its design emphasizes no model graph compilation step, small Docker images, fast boot times, token-based dynamic batching, optimized inference with Flash Attention, Candle, and cuBLASLt, Safetensors and ONNX weight loading, and production features such as OpenTelemetry tracing and Prometheus metrics.

Adoption history

Hugging Face's official deployment material places TEI inside the broader Inference Endpoints and embedding-container story. A Hugging Face blog on embedding endpoints presents Text Embedding Inference as the managed solution used to deploy open-source embedding models, and the SageMaker embedding-container announcement says the container is powered by TEI for efficient deployment of embedding models used in RAG applications.

How it is used

The normal package-nerd entry point is the text-embeddings-router executable or a ghcr.io/huggingface/text-embeddings-inference Docker image. Users select a Hugging Face model ID or local model directory with --model-id, expose HTTP endpoints such as /embed, /rerank, /predict, or OpenAI-compatible embeddings routes, and tune batch/request limits to match hardware.

Homebrew is explicitly documented for Apple Silicon local installs: the upstream README says users can brew install text-embeddings-inference and launch text-embeddings-router with Metal acceleration. Docker images cover CPU, CUDA architectures, ARM64, Hopper, Blackwell, and related hardware tiers.

Why package nerds care

TEI matters to package and infrastructure nerds because it turns a fast-moving ML serving stack into a versioned binary/container artifact. It pulls together model formats, GPU capability constraints, batching limits, metrics, tracing, Hugging Face Hub model IDs, private model tokens, and platform-specific acceleration.

Its Homebrew formula is notable because it gives Mac users a native local embedding server path outside Docker, useful for development, local RAG experiments, and testing Hub-compatible embedding models on Apple Silicon.

Timeline

• 2023: Hugging Face blog material presents Text Embedding Inference in embedding-model deployment workflows.
• 2024: Hugging Face announces a SageMaker embedding container powered by TEI for embedding models and RAG applications.
• 2025-2026: official docs and repository list expanded model families, hardware images, ONNX loading, OpenAI-compatible routes, Homebrew installation, and continued releases.

Related projects

• Hugging Face Hub supplies model IDs, revisions, private/gated model access, and compatible model tags.
• Text Generation Inference is the related Hugging Face serving project for generative language models.
• Candle, Safetensors, Flash Attention, ONNX, and cuBLASLt are cited upstream as core performance or loading technologies.
• MTEB and embedding model families such as BGE, E5, GTE, Nomic, Qwen, Jina, and Snowflake Arctic shape the models TEI users package and serve.

Sources

• https://github.com/huggingface/text-embeddings-inference - official README documents TEI goals, features, Docker usage, CLI, supported models, and Homebrew install.
• https://huggingface.co/blog/inference-endpoints-embeddings - official Hugging Face blog explains embedding deployment with Text Embedding Inference.
• https://huggingface.co/blog/sagemaker-huggingface-embedding - official Hugging Face blog announces an embedding container powered by TEI.
• https://huggingface.co/docs/text-embeddings-inference/index - official docs summarize TEI features and production deployment focus.
• https://huggingface.co/docs/text-embeddings-inference/quick_tour - official quick tour documents Docker deployment, /embed, /rerank, /predict, batching, and air-gapped usage.
• https://huggingface.co/docs/text-embeddings-inference/supported_models - official docs list supported model families and hardware images.