Achieve precise Knowledge retrieval with flexible metadata filters

We introduced a metadata-based filter DSL for Knowledge searches, enabling precise, policy-aligned retrieval at scale. You can now combine filters using EQ, IN, GT/LT, NOT, AND, and OR to target exactly the content you need. Initial support is available for PGVector-backed Knowledge bases. This reduces noise, improves result relevance, and helps enforce governance (for example, by segment, tenant, or timeframe) without complex post-processing.

Details

Compose boolean and relational filters to narrow results by metadata
Improves accuracy and reduces retrieval overhead in large Knowledge stores
Available with PGVector; pass knowledge_filters to opt in

Who this is for: Platform teams and builders managing large or multi-tenant Knowledge bases who need precise, metadata-aware retrieval and stronger guardrails.