𝙼𝚊𝚛𝚌𝚎𝚕 🥁

@trq212 How does it perform towards the further end of the context window, the same as at the beginning?

i think we might have undersold 1M context tbh, the performance is so so good, I really just don't clear the context window much these days

" ↑ Opus now defaults to 1M context · 5x more room, same pricing." - happy Sunday!

Just benchmarked my self-hosted LLM (Qwen3-32B on RTX 5090 + vLLM) with 30 concurrent nutrition questions. Results: → 30 requests completed in 0.87s total → 283 tokens/sec aggregate throughput → 93% accuracy (28/30 correct) → Avg latency: 0.44s per request vLLM's continuous batching is no joke: all 30 hit the GPU simultaneously and came back nearly at the same time. Self-hosting keeps getting more viable. #ai #selfhosted #localai

@swissspidy @jameswlepage Kind of my project this year for the @cloudfest this year. 😁 But with a different purpose (?). hackathon.cloudfest.com/project/wp-age…

@jameswlepage Kinda like my hackathon project last year :) These days, why not just use Claude Code or Gemini CLI woth WP-CLI? This can't possibly compete with them.

🤔 github.com/Automattic/wp-…

AI coding assistants can write WordPress plugins but can't test any of it. They iterate blindly. Just shipped wp-playground-mcp: gives AI assistants ephemeral WordPress instances on demand. Real WP, WebAssembly, boots in seconds. No Docker, no MySQL. Boot → Test → Tear down. Completes my agency's Pluginslab trifecta: 1. wp-devdocs-mcp → verified hooks for code 2. wp-blockmarkup-mcp → verified blocks for content 3. wp-playground-mcp → real WordPress for testing Author → Validate → Test. One prompt, tested plugin. github.com/pluginslab/wp-… #WordPress #MCP #AI #OpenSource #WebAssembly #WPDev #DevTools #ClaudeAI

wp-devdocs-mcp v1.1.0 is out, now indexes WordPress documentation too. Your AI assistant can query 7,200+ verified hooks AND 1,144 handbook pages from the Plugin Handbook, REST API, WP-CLI, Block Editor docs, and more. One command. Under a minute. No more hallucinated hooks or guessed answers. github.com/pluginslab/wp-… #WordPress #AI #OpenSource #WooCommerce #MCP #ClaudeCode #WebDev

@steipete Attending the Porto one. Also, have clients already that have a purchased bundles I created around openclaw that I’ll write a case study about soon. This is literally exploding.

This is humbling and insane and amazing. 🫶🦞luma.com/claw

Two MCP servers now cover the two biggest things AI gets wrong with WordPress: wp-devdocs-mcp → hooks & filters wp-blockmarkup-mcp → block markup Both work with Claude Code, Cursor, or any MCP client. Setup takes 2 minutes. github.com/pluginslab/wp-…

It parses block.json, save.js, edit.js and variations.js using AST analysis, generates markup examples, and validates them against the real save function. 121 core blocks, 171 WooCommerce blocks, 966 attributes, 520 validated markup examples,all queryable by your AI assistant via MCP.

Just open-sourced wp-blockmarkup-mcp,a companion to yesterday's wp-devdocs-mcp. AI assistants hallucinate Gutenberg block markup. Wrong attributes, wrong class patterns, wrongblock names. Every mistake = "Attempt Block Recovery" in the editor. This fixes it. Index the actual source code, give your AI verified data.

The fix: feed the LLM real data. Index WordPress Core → 3,400+ verified hooks Add WooCommerce → 3,500+ more Add Gutenberg, ACF, or your own private plugins Works with Claude Code and any MCP-compatible AI tool. 🔗 github.com/pluginslab/wp-… #WordPress #AI #OpenSource #WooCommerce #MCP

The problem: AI models generate hook names from memory. Most of the time they're right, but they also invent hooks that don't exist, use deprecated namespaces, and miss newer APIs like WooCommerce HPOS hooks. You only find out when the code doesn't work.

I just open-sourced wp-devdocs-mcp, a tool that stops AI coding assistants from hallucinating WordPress hook names. It indexes every action, filter, and JS API call from the actual source code and lets your AI query it before writing code. No more guessing. No more phantom hooks.

@chrislema 👀

This is the way.

@steipete you know when you have made a difference when Elon takes time to comment someone’s efforts to innovation. I’d take this as a complement.

@thepatwalls @openclaw /verbose on

To me, this feels like a fundamental flaw in @openclaw. If it's doing something in the background, or working on a longer-term project, I should at least be able to see this somewhere and maybe get some insights into what it's doing. Am I thinking about this wrong?

@AntonioCunha79 @XMuttley Se não fossem os pilotos, a Sata não teria a vida tão fácil. Sacrificam muito. E são altamente formados e experientes nestas situações.

@XMuttley Tem sim senhor. Fui aos acores este verao e foi 5 estrelas na sata.

Avião da SATA descola no meio da tempestade. Um A321 Neo com direção a Boston abana por todo o lado... dassss

@marinusklasen Opencode + Claude Opus 4.5

@schmitzoide What's your coding setup like? Are you using something like Auto Claude?

Have configured 3 different #clawdbot instances this afternoon with different roles. The world has changed forever. The way we all work now and specially write code has fundamentally changed. This is the first time I can clearly see how people are going to get replaced quicker then I thought.

@marinusklasen @jumperz For those it goes through a local model running on an RTX 5090.

@schmitzoide @jumperz Mail and calendar, does that mean all mail goes through claude (or other ai) servers?

Didn’t try. Will tomorrow. Run @openclaw on a Synology NAS server. It’s the closest to my backups and where all my files are from all the devices I have. Should be doable. 🙄