Ryan Dahl launched celld, an open-source implementation of Cloudflare Workers and Durable Objects that runs on self-hosted machines using V8, SQLite, and S3-compatible storage. Each Durable Object gets its own SQLite database with replication, eliminating the need for separate database clusters or consensus services. Version 0.1.0 is available under Apache 2.0 license in the denoland/celld repository.
Effect 4.0 is a major release that rebuilds the framework from scratch, delivering significant performance improvements, reduced memory usage, and smaller bundle sizes. The core package now has zero runtime dependencies and includes comprehensive features from basic functions to distributed systems with durable workflows and clustering. Starting with version 4.x, Effect offers long-term support with bug fixes until September 2029 and security fixes through 2031.
Comma's Salix distributed harness enables always-on agents to run persistent eBPF programs that monitor events and filter them before escalating to the main agent loop, avoiding expensive model calls on every tick. Agents write small C programs compiled to eBPF that execute 24/7 on shared cluster nodes with minimal memory overhead and safety guarantees. Spinfoam, a userspace eBPF runtime written in Rust, executes these Loops asynchronously without requiring kernel privileges.
Google has released Spanner Omni, a deploy-anywhere version of its distributed multi-model database that combines SQL, graph, vector search, and analytical capabilities. The platform, which has already achieved over 2 million downloads since its preview launch, enables enterprises to deploy Spanner across on-premises data centers, multiple clouds, and Kubernetes environments while maintaining Google-grade consistency and availability for AI workloads.
A semantic fuzzer for Obsidian Sync that automatically detects data-loss bugs by testing concurrent note edits across multiple clients. Built entirely with Claude Code artifacts (except the README), the project succeeded in finding reproducible sync failures but revealed significant limitations in using Claude for sustained software development without tight scope control.
A technical analysis of running PostgreSQL over QUIC (UDP) instead of TCP to overcome connection pooling bottlenecks in distributed systems. The author's experiment showed QUIC achieving 9,718 queries/second at 246ms latency versus TCP's 7,088ms latency under identical heavy load, demonstrating how latency asymmetry between clients and database clusters creates a transport bottleneck that connection pooling cannot solve.
nfltr extends Claude Code to run distributed agents across multiple machines using a hub-and-node architecture. The system allows work to be spawned on remote machines, persists when the hub disconnects, and gives each machine fine-grained control over what agents can do via opt-in capabilities.
Blob-stream is a Kafka alternative designed for low-cost, high-volume streaming that eliminates cross-availability zone network traffic, uses stateless brokers with no local storage, and integrates natively with Kubernetes. It achieves significant cost reductions and operational simplicity by diverging from Kafka's architecture while maintaining similar producer-consumer semantics and topics/partitions abstractions.