What is Weaviate?What is Weaviate?
AI-native vector database with hybrid search and built-in model integration
Weaviate in 60 seconds
An open-source, AI-native vector database that stores objects and their embeddings together and searches them by meaning.
Teams use it to power semantic search and RAG, with hybrid keyword-plus-vector ranking and embeddings built in.
- +Hybrid search blends BM25 keyword scoring with HNSW vector matching
- +Built-in vectorizer modules embed your data on import, no glue code
- +Stores objects and vectors together with filters on structured fields
- −Not a system of record: no ACID transactions and no relational joins
- −HNSW indexes sit in memory, so big datasets need real RAM budgeting
- −GraphQL-first query API has a steeper curve than plain SQL for newcomers
At a glance
| Category | Vector |
| Pronunciation | Pronounced "wee-vee-eight," a play on "weave" plus the "-ate" suffix. |
| First released | 2019 |
| Latest release | 1.39.0 (Aug 2026) |
| License | BSD-3-Clause |
| Written in | Go |
| Runs on | Linux, Macos, Windows |
| Deployment | Self-hosted, Managed, Embedded |
| Wire protocol | http, grpc |
| Query dialect | graphql |
| Consistency | eventual |
| ACID support | no |
| JSON support | native |
| Full-text search | native |
| Vector support | native |
| HA model | raft |
| Managed by | weaviate-cloud |
What is Weaviate?
Weaviate is an open-source vector database built for AI applications. Instead of storing rows and running exact-match queries, it stores objects together with their vector embeddings and finds the ones closest in meaning to a query. You load JSON data into collections, Weaviate turns each object into a vector, and a search for "laptop" can surface a document about "notebook computers" even when the words never match. It also keeps the structured fields alongside each object, so you can filter by price or category while ranking by similarity.
The project started in Amsterdam and had its first open-source release in 2019. It is developed by Weaviate B.V., the company formerly known as SeMI Technologies, founded by Bob van Luijt, Etienne Dilocker, and Micha Verhagen. Weaviate is written in Go and licensed under the permissive BSD-3-Clause license. The company follows an open-core model: the core engine is free to self-host, and Weaviate Cloud offers the same engine as a managed service.
Adoption climbed as retrieval-augmented generation became the standard way to ground large language models in private data. Weaviate now backs search and AI features at companies like Morningstar, Instabase, Stack AI, Neople, and Finster AI, and its own case studies show clusters serving millions of queries. The appeal is a single engine that handles embeddings, keyword search, and vector search without stitching several systems together.
How Weaviate works
Weaviate runs as a standalone server you talk to over REST, GraphQL, and gRPC. When an object arrives, a vectorizer module turns it into an embedding, either by calling a model provider like OpenAI or Cohere or by using vectors you supply yourself. The object and its vector are stored together, and the vector is added to the index that makes similarity search fast.
That index is HNSW, a graph of vectors where each search walks from node to node toward the nearest neighbors instead of scanning everything. HNSW is approximate, trading a little recall for a large speed gain, and it lives in memory, which is why RAM planning matters at scale. Weaviate also keeps an inverted index for BM25 keyword scoring, so a hybrid query can run both and fuse the two rankings into one result set.
For durability and scale, Weaviate persists data with an LSM-tree storage engine and shards collections across nodes. A Raft consensus layer manages schema and cluster membership, and replication keeps copies of each shard on multiple nodes so the cluster survives a node loss. Multi-tenancy isolates each tenant into its own shard, which is how single clusters serve thousands of separate users without mixing their data.
Key concepts
Collections and objects
Data lives in collections, the rough equivalent of a table, and each object is a JSON record with typed properties plus one or more vectors. A collection defines its properties, its vectorizer, and its index settings up front. Objects carry both the structured fields you filter on and the embeddings you search by, kept together in one place.
Vectorizer modules
Weaviate embeds data through pluggable modules. A text2vec module calls a provider such as OpenAI, Cohere, or Hugging Face at import time, or you can bring your own precomputed vectors. Because embedding is built in, the same query text is vectorized the same way on write and on read, which keeps results consistent without extra glue code.
HNSW vector index
Similarity search runs on an HNSW graph, where each query walks the graph toward its nearest neighbors rather than scanning every vector. The index is approximate, trading a little recall for a large speed gain, and it is held in memory. Quantization options like product and binary compression shrink that footprint when datasets grow large.
Hybrid search
Weaviate keeps an inverted index for BM25 keyword scoring alongside the vector index. A hybrid query runs both, then fuses the two ranked lists with a weighting you control through an alpha value. This catches exact terms like product codes that pure vector search can miss, while still ranking by meaning for everything else in the query.
Filtering and multi-tenancy
Because structured properties sit next to each vector, you can filter by price, date, or category in the same request that ranks by similarity. Multi-tenancy takes this further by isolating each tenant into its own shard. One cluster can host thousands of tenants with their data kept fully separate, the common pattern for SaaS applications.
Replication and sharding
Collections shard across nodes so a dataset can outgrow a single machine. A Raft consensus layer manages the schema and cluster membership, while replication keeps copies of each shard on several nodes. Reads can be served from any replica, and the cluster keeps answering queries when a node drops, with tunable consistency per request.
Weaviate by the numbers
Live GitHub adoption, updated daily
Who uses Weaviate
A handful of the companies running it in production
When to use Weaviate
Best for
Search that matches meaning rather than exact words, so a query finds relevant documents even when the phrasing differs. Built-in embeddings and HNSW make this the core job Weaviate is designed for.
Grounding a language model in your own data by retrieving the most relevant chunks for each prompt. Hybrid search plus generative modules let Weaviate fetch context and call the model in one pipeline.
Finding items similar to a product, image, or user profile by comparing their vectors. Weaviate handles text, images, and other media through the matching vectorizer modules for each data type.
Serving many customers from one cluster with each tenant isolated in its own shard. This keeps per-tenant data separate while sharing infrastructure, the usual pattern for AI features inside a SaaS product.
Not ideal for
Weaviate has no ACID transactions and is meant to index data whose source of truth lives elsewhere. Keep canonical writes in Postgres or a similar database and sync objects into Weaviate for search.
There are no multi-table joins, group-bys, or heavy aggregations here. For reporting, rollups, and column scans over billions of rows, a relational or analytics engine fits that work far better.
The HNSW index sits in memory, so a corpus that dwarfs your RAM budget gets expensive fast. Quantization and disk-based indexes help, but memory-bound scaling is still the main cost to plan around.
If you only ever fetch rows by an exact key or a plain filter, a vector database is overkill. A key-value store or relational table answers those lookups with less operational weight and less memory.
Weaviate vs alternatives
Head-to-head specs against the top 5 alternatives
| Spec | ||||||
|---|---|---|---|---|---|---|
| Identity | ||||||
| License | BSD-3-Clause | Proprietary | Apache-2.0 | Apache-2.0 | Apache-2.0 | Apache-2.0 |
| First released | 2019 | 2021 | 2021 | 2019 | 2022 | 2021 |
| Capabilities | ||||||
| ANN algo | HNSW | proprietary | hnsw | hnsw, ivf, diskann, scann | hnsw | — |
| Hybrid search | Native | Native | Native | Native | Native | Native |
| Vector | Native | Native | Native | Native | Native | Native |
| Ecosystem | ||||||
| Managed providers | 1 | 1 | 1 | 1 | 1 | 3 |
| Integrations | 4 | 4 | 11 | 7 | 4 | 7 |
| Use cases | ||||||
| Best for | Semantic search, RAG pipelines, and AI-native applications requiring hybrid vector and keyword search | Production-scale vector search with zero infrastructure management and enterprise security requirements | Semantic search, RAG pipelines, recommendation engines, image similarity, and AI agent memory with advanced filtering | Large-scale vector similarity search, RAG applications, and AI-powered recommendations | AI/LLM applications, RAG pipelines, semantic search, and rapid prototyping of embedding-based apps | Full-text search, log analytics, observability, and security analytics at scale |
| Not ideal for | Traditional relational workloads, complex transactions, or use cases requiring strong ACID guarantees | Self-hosted deployments, on-premise requirements, cost-sensitive prototyping, or workloads needing open-source flexibility | Traditional relational queries, OLTP workloads, time-series data, or use cases not involving vector embeddings | Traditional relational data, OLTP workloads, or applications not using embeddings | General-purpose data storage, OLTP, analytics, or production workloads requiring high availability | OLTP workloads, strong consistency requirements, or relational data modeling |
Quick start
Run Weaviate with Docker, connect with the Python client, create a collection, and run your first semantic search. A few minutes end to end.
docker run -p 8080:8080 -p 50051:50051 \
cr.weaviate.io/semitechnologies/weaviate:1.39.0pip install -U weaviate-clientimport weaviate
from weaviate.classes.config import Configure
client = weaviate.connect_to_local()
client.collections.create(
"Article",
vectorizer_config=Configure.Vectorizer.text2vec_openai(),
)articles = client.collections.get("Article")
articles.data.insert({"title": "A guide to notebook computers"})
results = articles.query.near_text(query="laptop", limit=3)
for obj in results.objects:
print(obj.properties)
client.close()The search for "laptop" finds the article about notebook computers because Weaviate matches on meaning. Swap in your own vectorizer or bring precomputed vectors, or browse and manage collections visually in a GUI.
Resources
Official
Related database guides
Qdrant
High-performance open-source vector database for next-generation AI applications
Milvus
High-performance cloud-native vector database built for scalable similarity search and AI applications
Chroma
Open-source AI-native vector database for building LLM-powered applications with embeddings
PostgreSQL
The world's most advanced open-source relational database
MongoDB
The most popular document database for modern applications
DuckDB
Fast in-process analytical database with rich SQL support and zero dependencies
Frequently asked questions
What is Weaviate used for?
Is Weaviate free and open source?
Is Weaviate a vector database?
Who makes Weaviate and where is it from?
How is Weaviate different from a keyword search engine?
Skip the config files
Connect to Weaviate in 30 seconds. Browse tables, run queries, and edit rows visually, on localhost, self-hosted, or cloud.
Open Weaviate in 1bench