NetMind.AI

While Everyone Was Chasing Claude Code's Hidden Features, We Turned the Leak Into 4 Practical Technical Docs You Can Actually Learn From After reading through a lot of the existing coverage, we found that most posts stopped at the architecture-summary layer: "40+ tools," "QueryEngine.ts is huge," "there is even a virtual pet." Interesting, sure, but not the kind of material that gives advanced technical readers a real understanding of how Claude Code is actually built. That is why we took a different approach. We are not here to repeat the headline facts people already know. These writeups are for readers who want to understand the system at the implementation level: how the architecture is organized, how the security boundaries are enforced, how prompt and context construction really work, and how performance and terminal UX are engineered in practice. I only focus on the parts that become visible when you read the source closely, especially the parts that still have not been clearly explained elsewhere. We published my 4 docs as downloadable pdfs in the link, but below is a brief. The Full Series: 1. Architecture — entry points, startup flow, agent loop, tool system, MCP integration, state management 2. Security — sandbox, permissions, dangerous patterns, filesystem protection, prompt injection defense 3. Prompt System — system prompt construction, CLAUDE.md loading, context injection, token management, cache strategy 4. Performance & UX — lazy loading, streaming renderer, cost tracking, Vim mode, keybinding system, voice input Overall The core is a streaming agentic loop (query.ts) that starts executing tools while the model is still generating output. There are 40+ built-in tools, a 3-tier multi-agent orchestration system (sub-agents, coordinators, and teams), and workers can run in isolated Git worktrees so they don't step on each other. They built a full Vim implementation. Not "Vim-like keybindings." An actual 11-state finite state machine with operators, motions, text objects, dot-repeat, and a persistent register. In a CLI tool. We did not see that coming. The terminal UI is a custom React 19 renderer. It's built on Ink but heavily modified with double-buffered rendering, a patch optimizer, and per-frame performance telemetry that tracks yoga layout time, cache hits, and flicker detection. Over 200 components total. They also have a startup profiler that samples 100% of internal users and 0.5% of external users. Prompt caching is a first-class engineering problem here. Built-in tools are deliberately sorted as a contiguous prefix before MCP tools, so adding or removing MCP tools doesn't blow up the prompt cache. The system prompt is split at a static/dynamic boundary marker for the same reason. And there are three separate context compression strategies: auto-compact, reactive compact, and history snipping. https://lnkd.in/et2GgJNU

Our CEO Kai Z., Product Lead Xiangpeng Wan, & Research Partner Dr Meng Fang from University of Liverpool were at Church House Westminster for the 𝟮𝟬𝟮𝟲 𝗔𝗜 & 𝗥𝗼𝗯𝗼𝘁𝗶𝗰𝘀 𝗥𝗲𝘀𝗲𝗮𝗿𝗰𝗵 𝗔𝘄𝗮𝗿𝗱𝘀 𝗖𝗲𝗿𝗲𝗺𝗼𝗻𝘆! Members of the team who couldn’t attend in person included Prof Fei Lu from John Hopkins University & Fei Xing from Mathematica. We were finalists for the Best Research Project (Industry Collaboration). This award recognises a research project that exemplifies outstanding collaboration with industry partners. It celebrates projects that foster mutual innovation, generate impactful co-created outputs, and translate research into real-world applications benefiting both academia and industry. Our project is "MathOdyssey: Benchmarking Mathematical Problem-Solving Skills in Large Language Models Using Odyssey Math Data" MathOdyssey is a collaborative benchmark project for evaluating mathematical and scientific reasoning in large language models, developed jointly by academia and industry partners. The project has achieved direct real-world impact through adoption by leading AI developers, including Google (Gemini 1.5) and Mistral AI (Mathstral). Building on the dataset, the team launched the Global Artificial Intelligence Championships, attracting 48 international teams, and organised the AGI Odyssey Symposium to foster sustained academia–industry dialogue. MathOdyssey demonstrates strong industry engagement, co-created outputs, and clear mutual benefit in advancing AI reasoning evaluation and deployment.

MiniMax M2.5 is the most popular model on NetMind right now, largely because it performs exceptionally well with OpenClaw, combined with very competitive pricing. And M2.7 has 88% win rate over M2.5! Want to use OpenClaw, which is blowing up, but not sure how to set it up properly? Check out our blog in the comments.

Our Product Lead, Xiangpeng Wan, visited the Lloyds Banking Group London office with the Cloud Platform Engineering London team by Ethan S.. Whilst the finance industry has long been a key client segment for our enterprise solutions, this was our first time participating in an on-site event at a financial institution. But this time, we didn’t just focus on business value, we also brought some fun. Xiangpeng introduced arena42.ai, where your agents can take part in games like debates, promotional challenges, or even meet partners in Agent Eden. And by winning these games, they can still earn you real rewards.

Honoured to be nominated!

First offline demo of NetMind XYZ’s social media agent in London! Last Wednesday, NetMind XYZ was introduced to the members of Trampoline NH C.I.C. We’ve built a social media agent tailored for small businesses in the F&B industry. Been disappointed by too many so-called social media agents? Here’s how our agent stands out: - Truly professional: built with real social media know-how - Sounds human: no more stiff, obvious AI copy - Truly agentic: it doesn’t need you to initiate every task manually - Built for restaurants: tailored workflows for F&B businesses - Real automation: you get a week's posts prepared in an hour Reboot your social media with us (link in comments!) Our 15% early bird discount is now live.

“After more than three years of hype since ChatGPT launched, the honeymoon period is over. Executives are desperate to realise the gains.” Our CCO Seena Rejal shared his perspective about a new study examining enterprise AI’s shift from experimentation to execution with Ricardo Oliveira in TechFinitive.com. Drawing from our recent work helping a fintech client transform millions of fragmented documents into a compliant RAG system, he emphasised that organisations must get their fundamentals right before deploying agentic AI effectively. As he noted: “You cannot scale what you cannot govern.” His contribution sits alongside perspectives from 18 other technology leaders. https://lnkd.in/d5pey5d6

Our CCO Seena Rejal included!

As posted previously, OpenClaw is super-trending in China and people are paying over $70 for house-call OpenClaw installation services. Tencent then organized 20 employees outside its office building in Shenzhen to help people install it for free. Their slogan is: OpenClaw Shenzhen Installation 1000 RMB per install Charity Installation Event March 6 — Tencent Building, Shenzhen Though the installation is framed as a charity event, it still runs through Tencent Cloud’s Lighthouse, meaning Tencent still makes money from the cloud usage. Again, most visitors are white-collar professionals, who face very high workplace competitions (common in China), very demanding bosses (who keep saying use AI), & the fear of being replaced by AI. They hope to catch up with the trend and boost productivity. They are like:“I may not fully understand this yet, but I can’t afford to be the person who missed it.” As a Reddit user commented under our post: The pressure from business leaders that think they know but don't know what they are requesting is leading their employees to do things that they don't know and can't control. This almost surreal scene would probably only be seen in China, where there are intense workplace competitions & a cultural eagerness to adopt new technologies. The Chinese government often quotes Stalin's words: “Backwardness invites beatings.” There are even old parents queuing to install OpenClaw for their children. How many would have thought that the biggest driving force of AI Agent adoption was not a killer app, but anxiety, status pressure, and information asymmetry? images from rednote