TL;DR
Three small but useful updates this week: (1) Weaviate 1.40.3 shipped with security bumps (CVE‑fixing deps), Go 1.26, and performance tidy‑ups in filters and REST Search; (2) Milvus tagged 3.0.4 (lake‑native) and 2.6.26 (LTS) for stability—release notes incoming; (3) Amazon Bedrock Knowledge Bases added a retrieval filter that can be applied per user query against a metadata schema, tightening control over filtered retrieval.
Weaviate 1.40.3: security and serving fixes on top of 1.40 GA
- Key facts and current state of the topic
- Weaviate 1.40.0 GA brought Namespaces, HFresh MUVERA, 4‑bit Rotational Quantization (RQ4), and a hardened Search REST API. Point releases are now hardening the line for production search. (github.com)
- Important context and background information
- Multi‑tenant, hybrid, and late‑interaction stacks are sensitive to replication, cache prefill/teardown, and module/dep CVEs; even “small” patches can stabilize p95/p99.
- Recent developments or changes
- 1.40.3 (Oct 29) includes dependency security bumps (e.g., go-jose v4.1.6), an upgrade to Go 1.26, and performance/operability fixes (e.g., batched Contains filter handling, REST Search polish). If you adopted 1.40.0/1.40.1, pull 1.40.3 for safer defaults. (newreleases.io)
Milvus: 3.0.4 (lake‑native) and 2.6.26 tags for stability
- Key facts and current state of the topic
- Milvus 3.0 GA introduced a lake‑native architecture (External Collections, Snapshots, Storage V3) and multi‑vector ergonomics; 2.6.x remains widely deployed as the stable line. (github.com)
- Important context and background information
- Teams operating hybrid (vector + scalar/text) pipelines rely on frequent point releases for stability (index loading, filtering, recovery paths).
- Recent developments or changes
- On Oct 26, tags for v3.0.4 and v2.6.26 were published (release notes “coming soon”), signaling ongoing hardening on both tracks. Plan canary rollouts once notes post, especially if you run heavy filtering or external‑table workflows. (github.com)
Amazon Bedrock Knowledge Bases: per‑query retrieval filters
- Key facts and current state of the topic
- Bedrock Knowledge Bases (GA June 2026) provide a managed RAG layer integrated with S3 Vectors and hybrid retrieval. (aws.amazon.com)
- Important context and background information
- Correct, explainable filtered retrieval is critical at scale (ads, marketplaces); platform‑level filtering reduces glue code and prevents silent recall loss.
- Recent developments or changes
- Doc history for Oct 26 notes a new “apply a retrieval filter based on a user query and a metadata schema” capability, letting you dynamically constrain results at request time. Evaluate for tenant/attribute scoping and compliance‑sensitive use cases. (docs.aws.amazon.com)