--- title: "Weaviate and RAG, where things stand" description: "A plain look at where vector databases and RAG actually stand right now, and why the database is the easy part." canonical: "https://www.pollyester.com/blog/weaviate-and-rag-where-things-stand" date: "2026-07-17" author: "Pouya Nafisi" --- [Pollyester](https://www.pollyester.com/index.md) › [Ideas](https://www.pollyester.com/blog.md) › Weaviate and RAG, where things stand # Weaviate and RAG, where things stand **A plain look at where vector databases and RAG actually stand right now, and why the database is the easy part.** ![Abstract silk-thread hero. A dense weave of turquoise and silver filaments passes through a plain, clearly interchangeable ring, the commodity vector store, while the real structure, the chunking, reranking, and measurement, is held in the intricate weave around it.](https://www.pollyester.com/blog/weaviate-and-rag-where-things-stand/banner.dark.webp) If you're building RAG right now, you'll hear a lot about vector databases. Weaviate, Pinecone, Qdrant, Milvus, and a dozen more. It's worth understanding what they do, and how much they actually matter, because the honest answer to the second question is "less than the noise suggests." Here's what a vector database is for. When you do retrieval, you turn your documents into embeddings, store them, and search by similarity. The vector database stores those embeddings and searches them quickly. Weaviate is one of the good ones. It's open source, it does hybrid search, meaning vector and keyword together, it handles multi-tenancy well, and its managed cloud is solid. What's interesting about Weaviate lately is that it's becoming more than a database. It's added Agents, layers on top for querying, transformation, and personalization. You can see the same move across the category. The database part is turning into a commodity, so the companies are building upward. And it really is turning into a commodity. Postgres does vector search now with pgvector. So do Mongo, Elastic, Redis, and the big clouds. For a lot of projects, the simplest path is to keep your vectors in the database you already run, and reach for something like Weaviate only when scale calls for it. The vector database usually isn't where a RAG system succeeds or fails, though. That comes down to the parts around it. How you chunk your documents. Whether you pair vector search with keyword search, since similarity alone struggles with names, part numbers, and exact phrases. Whether you rerank results before the model sees them. And whether you measure any of it, which is the step most people skip. There's also the question that keeps coming up, whether long context makes RAG unnecessary. Models read a million tokens now, so why retrieve? In practice, a million tokens is slow and expensive, and putting everything in context makes it hard to trace where an answer came from. So retrieval hasn't gone away. It's shifted, from a fixed pipeline into something the model reaches for as a tool when it needs it. So if you're starting out, Weaviate is a good choice, and pgvector is a good simpler one. Either way, the database is the easy part. Most of the work is everything around it. ## Related reading - [The Data Is the Moat. The Model Is a Rental.](https://www.pollyester.com/blog/data-is-the-moat-model-is-a-rental.md) - [The Model Is a Commodity. The Judgment Isn't.](https://www.pollyester.com/blog/model-is-a-commodity-judgment-isnt.md) - [Stop Doing GEO. Fix the Data Underneath It.](https://www.pollyester.com/blog/stop-doing-geo-fix-the-data-underneath.md) Older: [What Is an AI-Native Growth Agency?](https://www.pollyester.com/blog/what-is-an-ai-native-growth-agency.md) --- ## Pollyester — site navigation Primary: [Services](https://www.pollyester.com/services.md) · [Work](https://www.pollyester.com/work.md) · [Partners](https://www.pollyester.com/partners.md) · [About](https://www.pollyester.com/about.md) · [Ideas](https://www.pollyester.com/blog.md) Company: [About](https://www.pollyester.com/about.md) · [Careers](https://www.pollyester.com/careers.md) · [Contact](https://www.pollyester.com/contact.md) · [How a project runs](https://www.pollyester.com/process.md) · [Engagement](https://www.pollyester.com/engagement.md) Services: [Get found](https://www.pollyester.com/services/get-found.md) · [Get chosen](https://www.pollyester.com/services/get-chosen.md) · [Grow the customer](https://www.pollyester.com/services/grow-the-customer.md) · [Earn advocacy](https://www.pollyester.com/services/earn-advocacy.md) · [Pick, pack & ship](https://www.pollyester.com/services/pick-pack-ship.md) · [Order orchestration](https://www.pollyester.com/services/order-orchestration.md) · [Cost optimization](https://www.pollyester.com/services/cost-optimization.md) · [Build the agent-ready core](https://www.pollyester.com/services/build-the-agent-ready-core.md) · [Brand & identity](https://www.pollyester.com/services/brand-identity.md) · [Creative & art direction](https://www.pollyester.com/services/creative-art-direction.md) · [Content & campaigns](https://www.pollyester.com/services/content-campaigns.md) · [Video & media](https://www.pollyester.com/services/video-media.md) · [Storefront & product](https://www.pollyester.com/services/storefront-product.md) · [Experiential & activations](https://www.pollyester.com/services/experiential-activations.md) · [Events & retail](https://www.pollyester.com/services/events-retail.md) · [Design & build](https://www.pollyester.com/services/design-build.md) Legal: [Privacy](https://www.pollyester.com/privacy.md) · [Terms](https://www.pollyester.com/terms.md) Full index: [llms.txt](https://www.pollyester.com/llms.txt) — the markdown map of the site.