Weaviate 1.27+: Multi-Tenancy, Dynamic Indexes, and Generative SearchMixed sources12 sources, drawn from a mix of official docs, project repositories and research.This rates where the sources come from. It doesn't prove every claim.
By Amit UpadhyayPublished Updated 14 minArticle 5 of 13 in Databases
AI assistant · answers from this article
1 of 6
Purely conversational queries
Nothing to add yet
If every search is a question in ordinary words, matching by meaning is enough.
On this page
Weaviate, in one sentence
Weaviate is a vector database that scores keyword matches and meaning together in one hybrid query, so a search can find an exact code and a paraphrase alike.
You reach for it when searches mix natural-language questions with exact terms such as error codes and product names. Not before: it costs a system to run (or a managed plan), schema design up front, and the alpha weight to tune.
The cheat sheet
What to use when
You do not need all of this on day one. Each row is one signal and the piece that answers it. Stop at the row your system has reached, or pick one to see it in the picture above.
Where you areThe signalWhat to useWhat it costs
None. Nobody searches for an identifier.Nothing more. Vector search aloneNothing
“Customer getting E-4102 at checkout” finds articles about checkout and ignores the code.Hybrid search: keyword and vector scores in one queryStep 3 ↓A weight to choose between the two
Alpha is 0.5 because it sounded neutral, and nobody has measured it.Tune the weight on labelled queriesTesting hybrid quality ↓A labelled set to build and to rerun
A field was inferred as the wrong type, and filters on it do not work.Define collection properties and types explicitlyStep 1 ↓A schema to maintain
All tenants sit in one shard, and every small tenant carries a full index in memory.Built-in multi-tenancy, with the dynamic indexCommon mistakes ↓A feature still marked experimental
One node was easy. A replicated cluster means upgrades, memory pressure and shard placement.Decide on purpose between self-hosting and the managed serviceSelf-host or managed? ↓Operations work, or data outside your own network
Start With Vector Search Alone
If your users only ask conversational questions, vector search by itself is enough, and you don't need a second ranking. Hybrid search earns its place when queries also carry exact terms, such as an error code or a product name, that meaning-based search tends to ignore.
Support agents search with things like “customer getting E-4102 at checkout”. Vector search understands the second half and ignores the error code, so the one guide that mentions E-4102 ranks below general pages.
Weaviate combines keyword (BM25) and vector scores in one hybrid query, with a weight you can tune, so exact codes and paraphrases are both found. It is open source, can be self-hosted or used as a managed service, and has added multi-tenancy and dynamic indexes as tenants grow. Below is how to set it up and tune it on your own queries.
If vector search keeps missing codes and names
You need a vector DB you can run inside your own VPC
You have 10k tenants and Pinecone's pricing makes you wince
You want hybrid (BM25 + vector) in one query, not glued together in app code
You don't know the difference between a "collection" and a "tenant" in Weaviate
Pro deep dive14 min full read
Keep reading with Pro
This is one of our advanced production deep dives. Pro unlocks the rest of it, every other deep dive, and the See, Animate, Test and Build modes on every topic.