When building AI applications with Weaviate, good early architectural decisions can aid future development velocity. Some choices are expensive to change later - like vectorization strategies and schema design - while others can be adjusted as you scale.

This guide focuses on the "when" and "why" behind different architectural choices, organized as actionable decision patterns you can apply to your specific use case.

Database-level decisions

Database-level decisions affect your entire deployment and are among the hardest to change later. These choices determine operational overhead, scalability, and feature availability.

Deployment type

Weaviate's core database functionality is the same across deployment options. The main difference is operational overhead.

Two managed services are the exception: the Query Agent and Weaviate Embeddings run in Weaviate Cloud only and are not available in a self-hosted deployment. If your design depends on either, that constrains your deployment choice.

If you want self-service deployment → Use Shared Cloud

• Fully-managed SaaS on shared infrastructure, with all operational tasks managed through the web console
• Example: Startup focused on product development, not infrastructure

If you need the Query Agent → Use Weaviate Cloud

• The Query Agent runs agentic search over a Weaviate Cloud database: ask a question in natural language and it decides which collections to search, and which filters, sorts and search types to apply
• It is a Weaviate Cloud service, so it is not available to self-hosted deployments, but it is included on free clusters, with a monthly allowance
• Example: Natural language query interface without custom development

If you have a large dataset, or high throughput requirements → Consider Dedicated Cloud

• Dedicated Cloud is much better suited for provisioning custom, larger resources
• Example: SaaS company experiencing healthy growth and considering next steps

If you want a bespoke, but managed solution → Use Dedicated Cloud

• Dedicated instance on isolated infrastructure, with custom configurations managed by Weaviate
• Example: Large enterprise with specific networking, security, or compliance needs

If you must have full control, have DevOps capacity and resources → Consider Self-hosting

• You manage everything - infrastructure provisioning, updates, scaling, monitoring, disaster recovery, etc.
• Example: Where air-gapped environments are a must, or no external party involvement allowable

Older names you may still see

Weaviate updated its Cloud offering on October 27th, 2025. Shared Cloud clusters were previously called Serverless clusters, and Dedicated Cloud clusters were previously called Enterprise clusters. Older material may still use the retired names.

Learn more

• Installation: Deploying Weaviate | Weaviate Cloud: Create a cluster

What's next?

Once you've chosen your deployment type, you need to plan how your database will scale as your data and usage grow.