Weaviate

Weaviate is an open-source vector database that helps developers store, search, and retrieve data using vector embeddings for semantic search, AI applications, and RAG workflows.

At a Glance

Pricing Freemium

Weaviate is an open-source vector database for semantic search, recommendations, and retrieval-augmented generation (RAG). It stores records and their metadata alongside embeddings, which are numeric representations of content. This lets applications find related results even when the query uses different words. Developers can also combine vector search with keyword matching and filters, such as searching only published documents in a specific language.

How Weaviate Works

Weaviate uses vector indexes to find similar content and inverted indexes to handle keyword searches and metadata filters. A single query can use both.

  • Set Up a Collection: Define what each record will contain. A document collection might include a title, text, publication date, and language.
  • Add Embeddings: Import vectors you’ve already generated or connect a supported embedding model to create them as records are added.
  • Choose a Vector Index: HNSW is the default. Flat is suitable for small datasets, while Dynamic starts with Flat and switches to HNSW after the dataset reaches a configured size.
  • Run a Search: Use vector search for related concepts, BM25 for keyword matches, or hybrid search for both. Add filters to limit which records can appear.
  • Use Results in a RAG App: Pass the retrieved records to a language model so it can answer using your data. Anthropic, for example, is available as a text generation integration.

How We Rated Weaviate

We reviewed Weaviate’s official documentation and pricing, focusing on search options, filtering, index configuration, deployment, and tenant isolation. This was a documentation review. We did not benchmark query speed, memory use, or update performance.

Pros

  • Combines semantic search, keywords, and filters.
  • Returns record details with search results.
  • Supports multiple embeddings per record.
  • Separates data for different customers.
  • Offers self-hosted and managed cloud options.

Cons

  • Large HNSW indexes can use substantial memory.
  • Index settings take testing to tune.
  • Self-hosting puts infrastructure management on your team.
  • The free cloud plan allows one collection and up to three tenants.
  • Hosted embeddings and Query Agent usage can add costs.

  • Developers building chatbots that answer from company documents.
  • Ecommerce teams that need both product-name matches and searches by description.
  • SaaS teams that need to keep each customer’s records separate.
  • Search teams building text and image retrieval with compatible embedding models.
  • Engineering teams that want to choose between managing their own database and using a cloud service.

  • Find results by meaning while keeping exact terms, such as product codes, in the ranking.
  • Limit searches by language, category, or publication status.
  • Search the same record by title or description using separate embeddings.
  • Retrieve the source text and metadata a chatbot needs to answer a question.
  • Use the Cloud Query Agent to ask questions across collections in plain language.

Weaviate's Key Features

Stores records, metadata, and embeddings together.

Combines vector similarity with BM25 keyword search.

Filters results by fields such as category, date, or language.

Supports multiple named vectors for the same record.

Accepts existing embeddings or generates them through supported models.

Supports HNSW, Flat, Dynamic, and HFresh vector indexes.

Keeps each customer’s data separate with multi-tenancy.

Offers a Cloud Query Agent for natural-language questions.

Frequently Asked Questions

How does Weaviate handle hybrid search using vector and keyword queries?
Weaviate runs vector and BM25 keyword searches in parallel and combines their scores. The alpha setting controls how much weight each search gets. The default, relativeScoreFusion, combines normalized scores. The optional rankedFusion method uses each result’s position in the two rankings.
Can Weaviate store and retrieve multimodal data such as images and text?
Yes. With a compatible multimodal embedding model, Weaviate can index text and images so an application can find images from a text query. Supported media types depend on the model. A record can also have separate named vectors for different search needs.
How does Weaviate support metadata filtering during vector searches?
Add conditions on record properties to limit which objects are eligible for a match. For example, a search for similar product descriptions can return only items in a chosen category that are in stock. Weaviate uses its inverted index to speed up filtering.
What is the difference between self-hosted Weaviate and Weaviate Cloud?
With self-hosting, your team runs the database and manages infrastructure, upgrades, and backups. Weaviate Cloud provides a managed cluster, with backup, support, and deployment options that depend on the plan.
How can developers configure vector indexes for large-scale AI applications?
Choose an index type in the collection configuration. HNSW is a common starting point for larger datasets. Test compression and index settings with your own queries to see how they affect memory use, response time, and how often the search returns the right results. Each named vector can have its own index settings.
Does Weaviate support multi-tenancy for applications serving different customers?
Yes. When multi-tenancy is enabled on a collection, each tenant’s data lives in a separate shard. Reads and searches target a specific tenant, keeping that customer’s records separate from others in the same collection.

0.0

Based on user reviews

Reviews are moderated before they appear here. Share your experience with Weaviate to help others decide.

Write a review

For AI Builders

Built an AI Tool? Get It Listed.

Reach thousands of professionals actively hunting for new AI solutions every single day.