AI & Agent Frameworks
Articles on AI agent frameworks, open-source agent development tools, and platforms for building and deploying AI agents.
Four Agentic AI Memory Systems for Smarter LLMs
An overview of four third-party projects—Graphiti, Hindsight, Mem0, and Supermemory—that expand how AI agents and LLMs can draw on facts, documents, and conversations to deliver results through persistent memory capabilities.
The "Five-Character Fix": Decoding the Lighthouse Agentic Browsing Audit's llms.txt Requirement
A detailed analysis of how Chrome's new Lighthouse Agentic Browsing audit parses llms.txt, why failing the audit doesn't mean your content is bad, and how a simple formatting tweak enables a clean pass.
Vercel Ships Open-Source Agent Framework 'eve' and Tackles Shadow AI With Passport
At its Ship event in London, Vercel unveiled eve — an Apache 2.0 open-source agent framework built on TypeScript and Markdown with sandboxed VMs — alongside Passport, an OpenID Connect identity layer designed to bring employee-built AI apps under enterprise control.
Google’s New OKF and ARD Specs: What They Actually Do, and Why Your SEO Stack Shouldn’t Panic
Breaking down Google’s Open Knowledge Format and Agentic Resource Discovery specs — what they really solve, how they stack against MCP and LLMs.txt, and what you should actually do (hint: don’t rewrite your site in markdown).
The Accessibility Tree Decides Whether an AI Agent Can Read Your Page
AI agents don't read your website like humans do. They read the accessibility tree — a stripped-down structural model that powers screen readers and now drives autonomous agents. With 57.2% of HTML traffic now automated (Cloudflare, June 2026) and WebAIM reporting web accessibility regressed for the first time in six years, site owners face a new reality: the structure that lets machines understand your page is breaking down just as more traffic depends on it.
DSpark's 85% Speed Claim Hides the Real Bottleneck: Acceptance Quality
DSpark is an open-source orchestration framework developed by DeepSeek to accelerate LLM inference. While raw decoding speedups of up to 85% have been demonstrated, final real-world acceptance quality remains a critical factor in overall system performance.