LoginSign Up
← Back to all sparks
W

Weaviate

DEVOPS
Velocity5.0

Open-source vector database for AI applications and semantic search

weaviate.io ↗

Weaviate is competing on memory footprint while patching a credential leak.

vector-searchquantizationagent-memorysecurityrag
◆Current state
Weaviate's latest release is v1.39.3, a high-severity fix for credential disclosure in its Google modules. The surrounding entries are mostly engineering write-ups on what 1.39 shipped — 4-bit Rotational Quantization and the disk-based HFresh index — plus guides on Engram agent memory and multi-vector PDF retrieval.
◆Where it's heading
The technical thread is cost per vector: quantization and disk-resident indexes both cut RAM requirements for large collections. In parallel, Weaviate is building up agent-facing layers (Engram memory, the Query Agent's effort tiers) on top of the database.
◆Prediction
4-bit Rotational Quantization, previewed in 1.39, is the likeliest candidate for promotion to GA in the next minor release.

◆Recent moves

  1. 2d ago

    Weaviate security release - High severity fix for credential disclosure in the Google modules

    v1.39.3 fixes a high-severity credential disclosure issue in the Google modules. Deployments using Google embedding or generative integrations should upgrade promptly.

    View source ↗
  2. 11d ago

    Agent Memory with Engram: A Practical Guide

    A usage guide for Engram covering topic design, scopes, retrieval modes and prompt-cache-friendly placement. It documents the agent-memory layer rather than changing it.

    View source ↗
  3. 16d ago

    4-bit Rotational Quantization

    A deep dive into 1.39's 4-bit Rotational Quantization, including SIMD work, a centered tier and scaling analysis. It underlines that memory-per-vector is where Weaviate is pushing hardest.

    View source ↗
  4. 24d ago

    HFresh: Memory-Efficient Vector Search

    HFresh is a disk-based index with low heap use and incremental background maintenance. Paired with quantization, it lets larger collections run on smaller machines.

    View source ↗
  5. 25d ago

    Building Foundry Part 3: From archive to creative search

    Part 3 of a tutorial series applying hybrid search to a creative-asset archive. It is an example application, not a product change.

    View source ↗
  6. 1mo ago

    How to extract meaning from charts and tables in PDFs

    A tutorial on late-interaction multi-vector retrieval for chart- and table-heavy PDFs. It shows how existing multi-vector support can skip OCR and chunking.

    View source ↗