Three layers. One place to think, build, and share what we learn.
Choose a topic to explore AI agents, application architecture, runtime and rendering, or real product workflows. Browse twelve articles at a time, newest first.
Showing 1–12 of 35 articles
Latest articles
Ideas and practical guides across Semitexa OS, Platform, and Framework.
REVERB / SSE / ONE BOARDx2→One live board, built twice.
The same live order board built twice: Laravel with Livewire and Reverb, and Semitexa with Server-Sent Events. Code, measured latency, moving parts, and where each one is worse.
MCP replaced its HTTP+SSE transport with Streamable HTTP. What changed, when to use stdio, Streamable HTTP or plain SSE, and how Semitexa uses stdio for MCP and SSE for the browser.
How EventSource reconnects, what Last-Event-ID really does, and how a server replays missed SSE events or tells the page to re-sync. With two bugs we shipped.
Prompt management for PHP: prompt templates with ids, fail-closed rendering, a preview of what the model receives, per-tenant overrides and drift checks.
Harness engineering for PHP: guides and sensors around a coding agent, the PHP tools for each, running the right checks per change, and a Laravel starter.
Claude Code reads AGENTS.md since 2.1.277, but only without a CLAUDE.md. How to keep one instruction file for every agent, and how ai:verify keeps it short.
Follow one click from a fingertip through TLS, undersea fiber and a Swoole worker in Helsinki back to every open screen, then press a live Say hello button.
Laravel Octane, Hyperf and Semitexa compared: what each keeps in memory, where request state lives, tenancy, live pages, AI agents, and when to choose which.