Skip to main content
Go to documentation:
⌘U
Weaviate Database

Develop AI applications using Weaviate's APIs and tools

Deploy

Deploy, configure, and maintain Weaviate Database

Query Agent

Run agentic search over your Weaviate Cloud collections

Weaviate Cloud

Manage and scale Weaviate in the cloud

Engram

Persistent memory for LLM agents and applications

Additional resources

Integrations
Weaviate Academy

Need help?

Weaviate LogoAsk AI Assistant⌘K
Support
Community Forum
Contributor guide

Changelog

This page documents changes and notable updates to the Weaviate Query Agent. As the Query Agent is a managed service, improvements roll out automatically though sometimes you may need to upgrade the client to use new features.

September 2026​

Faster responses for long conversations​

The Query Agent now dynamically selects the optimal service tier based on conversation length, automatically upgrading requests to a faster priority tier where possible. Combined with recent model price reductions, this delivers quicker responses at a lower cost — especially for long multi-turn conversations. This happens automatically for all requests — see multi-turn conversations for how to make the most of conversational context.

July 2026​

New effort parameter in Search Mode​

Search Mode gained an optional effort parameter, letting you choose how much work the agent puts into each search. Higher effort levels widen retrieval for better recall and add an extra reranking pass for higher-quality results. Learn more in the Search Mode guide.

Upgraded to the latest OpenAI models​

The Query Agent was upgraded to OpenAI's latest Luna and Terra models, delivering improved results on our internal Search Mode benchmarks along with better query understanding across all modes.

Added automatic data exploration for filters​

Ask mode and search mode now have an automatic step which explores the structure of your data before making any queries. This helps to narrow down specific values for properties, so it can filter on exact strings or values without guessing.

Structured outputs in Ask Mode​

Ask Mode now supports structured outputs: provide an output_format schema and receive the final answer as a validated, typed object instead of free text. See the structured outputs reference.

Follow-up query suggestions​

Suggest Queries mode can now take a conversation history and suggest follow-up queries that continue the user's current thread, rather than generic queries about the collection. See the Suggest Queries guide.

May 2026​

Customizing the result evaluation​

You can toggle the degree of result evaluation in Ask Mode, allowing an LLM to look at the final answer and determine which sources were used to provide the answer (higher latency), or disabling this functionality (lower latency). See the parameters table.

Was this page helpful?