Weaviate clicked for me the moment I needed more than basic vector search. We were building a RAG pipeline for internal docs and kept hitting the same problem: pure vector search would return semantically similar results but miss exact keyword matches. Weaviate's hybrid search solved that in one query, mixing vector similarity with BM25 scoring. First time it worked I literally said "why isn't this standard everywhere?" The module system is great too, swapped from OpenAI embeddings to Cohere without changing a single line of application code.
That said, the learning curve is real. Schema-first means you have to plan your data model upfront before indexing anything, which slows down the "just throw data in and see what happens" phase. If you're used to Pinecone's simplicity where you just push vectors and go, Weaviate will feel heavier at first. Self-hosting also requires some infra knowledge. We burned a weekend getting the Docker setup right with proper resource allocation.
For the price though it's hard to beat. Self-hosted is completely free under BSD-3, and the managed cloud starts at $45/month which is reasonable for what you get. If your use case involves structured data where metadata matters as much as semantic similarity (product catalogs, legal docs, enterprise knowledge bases), Weaviate is genuinely one of the best options out there right now. Just budget some ramp-up time for your team.
Tesla Reader
Finally, a tool that makes AI development components this accessible. Open source and built on solid foundations too—definitely going to give it a try.