Welcome to Embedded Analytics

Anthropic is suspending access to its Fable 5 and Mythos 5 models following a national security directive from the US government. The government expressed concerns about potential security bypasses, though Anthropic argues the reported vulnerabilities are minor and common in other models. The company plans to comply with the order while working to restore access as soon as possible.

Anthropic has introduced hidden safeguards that can secretly limit Claude's effectiveness when users ask about AI development. The author warns that this creates a supply chain risk because developers cannot tell if the model is failing due to a technical error or a hidden policy. This lack of transparency makes it difficult for companies to fully trust the tool for their software development.

MariaDB has introduced a new DuckDB storage engine that allows users to run high-speed analytical queries directly on the same server as transactional data. This integration enables users to perform complex joins between different data types without needing separate systems or ETL pipelines. It is designed for hybrid transactional and analytical processing, offering faster performance for large-scale data analysis while maintaining a familiar SQL interface.

The article explains how Momentic migrated its caching system from Postgres to ClickHouse to handle massive data growth. By using ClickHouse's sparse primary index and materialized views, the team reduced latency and improved scalability. This transition allowed them to process billions of cache entries daily with high performance.

S&P Dow Jones Indices has denied SpaceX's request for fast-tracked entry into the S&P 500. This decision prevents the company from receiving billions in automatic investments from passive funds due to its lack of profitability and high debt. The ruling also means that AI companies like OpenAI and Anthropic will not receive similar accelerated access to the index.

The article describes a lightweight logging stack using Fluent Bit, Parquet files, and DuckDB for small environments. It explains how to efficiently store logs in S3, use Hive partitioning for better performance, and query them using SQL in Grafana. The author also highlights techniques like buffering, aggregation, and compaction to manage file sizes and costs.

ClickHouse has significantly improved its join performance over the last two years, making it 26 times faster on heavy workloads. The updates include better query planning, lazy column replication, and automatic join reordering based on statistics. These engineering improvements allow the database to handle complex SQL queries more efficiently with default settings.

This article explains why using NULL values in ClickHouse can lead to slower performance compared to using default values like zero. Because ClickHouse is a columnar database, it manages NULLs using a separate byte map that adds extra processing and overhead. The author recommends avoiding NULLs in production to maintain better speed and efficiency.

This article explains the benefits and challenges of using a table-per-tenant approach in ClickHouse. It provides practical advice on managing parts, handling server startup times, and preventing schema drift. The author also recommends using dedicated nodes and proxy routing to balance read and write workloads effectively.

DuckDB version 1.5.3 introduces several new features for Iceberg tables, including MERGE INTO support and the ability to use ALTER TABLE for schema evolution. The update also adds support for bucket and truncate partition transforms and the latest Iceberg v3 specifications. These improvements allow users to perform more complex data operations and manage metadata more efficiently within the DuckLake ecosystem.

The authors announced new features for OpenData Buffer that allow high-volume logs to be sent to ClickHouse using S3 as a storage layer. This setup achieves 1.1 Gbps of throughput for around $180 per month, which is significantly cheaper than using traditional systems like Kafka. The new pipeline uses parallel processing and stable batch identities to ensure fast, reliable, and cost-effective data delivery.

This article explores DuckDB's new vector similarity search (VSS) and the Quack protocol for client-server communication. The author demonstrates how these features can work together to route and search data across different database instances. The post includes a practical example using image embeddings to show how DuckDB can handle complex, distributed data tasks.

ClickHouse Cloud has launched a beta service for managed Postgres that uses NVMe storage for faster transactions. This platform allows users to sync data automatically into ClickHouse for real-time analytics and use a unified query layer. The service is currently free until June 2026, with a 50% discount offered during the beta period.

ClickHouse Cloud has introduced multi-stage distributed query execution to better handle large-scale data workloads. This new system allows intermediate data to be repartitioned between execution stages, removing bottlenecks in large joins and high-cardinality aggregations. Testing shows significant performance gains, with some queries running up to 7.4 times faster on multiple nodes.