Self-hosted embedding service stacks
Turn text into vectors on your own infrastructure, for semantic search, clustering and RAG retrieval.
Embedding stacks
- Embedding Service (embedding) – High-performance text embedding service with caching
About Embedding stacks
An embedding service exposes an open embedding model over HTTP so other services can turn text into vectors. Self-hosting it removes per-token API costs for large re-indexing jobs and keeps the text you embed private.
Points that matter in practice:
- The embedding dimension fixes the schema of your vector index; changing models means re-embedding everything
- Many models expect a query or passage prefix, and leaving it out quietly hurts retrieval
- Batching and a GPU make bulk indexing much faster, while CPU is often fine for query-time traffic
Frequently asked questions
- What is an embedding service?
- A small API that converts text into numeric vectors with an embedding model. Vector databases store these vectors so that similar texts can be found by meaning rather than keywords.
- Can embedding models run on CPU?
- Yes. Most embedding models are small enough for CPU inference. A GPU mainly helps when you embed large document collections in bulk.
- What happens if I switch embedding models?
- Vectors from different models are not compatible, so you have to re-embed all documents and rebuild the index with the new model's dimension.
Guides
- How to Self-Host a RAG Pipeline with Docker Compose – A complete, working self-hosted RAG stack in one docker-compose.yml: Ollama for chat and embeddings, Qdrant for vectors, and a small Python service for ingestion and queries.
- pgvector vs Qdrant vs Weaviate vs Milvus vs Chroma for RAG – A practical comparison of the five vector databases people actually self-host for RAG, with a decision guide by scenario instead of benchmark charts.
- How to Choose the Best Embedding Model for Self-Hosted RAG – A practical way to choose a self-hosted embedding model: what the leaderboard can and cannot tell you, what each dimension costs in storage, and how to serve, swap and rerank.