Weekly Claws #4: The New Computer

“Every company in the world today needs an OpenClaw strategy.”

That’s Jensen Huang at GTC, calling OpenClaw “the new computer” and comparing its impact to Linux reshaping the software stack. OpenClaw became the most popular open-source project in NVIDIA history within weeks of launch.

This isn’t just an endorsement — it’s a historical claim. NVIDIA didn’t just build tools for agents. They’re building the entire enterprise stack around OpenClaw.

This week: purpose-built agent hardware, local AI tools going mainstream, the first agent-only chess tournament, and the infrastructure layer finally catching up to what agents actually need.

Let’s dive in. 🐾

🚀 Releases & News

The Big One

🔥 Jensen Huang at GTC: OpenClaw Is “The New Computer” — NVIDIA CEO calls OpenClaw essential for every company, comparing its impact to Linux reshaping the software stack. OpenClaw became NVIDIA’s most popular open-source project within weeks. New hardware (Vera CPU), new stack (NemoClaw), and direct DGX Spark integration — they’re not just supporting OpenClaw, they’re building the future around it.
🔒 NVIDIA’s Version of OpenClaw Could Solve Its Biggest Problem: Security — NVIDIA isn’t just adopting OpenClaw — they’re hardening it. Own fork with security-first approach for enterprise. Sandboxed environments, policy-enforced access, and the infrastructure for always-on agents that don’t leak your data.

OpenClaw Updates

🐦 A packed week of releases for OpenClaw — Three in four days (2026.3.11–3.13): 1M context models via OpenRouter, GPT-5.4 fixes, Gemini Embedding 2, Dashboard v2, /fast mode, live Chrome sessions, Android down to 7MB, 12+ security patches. Plus: OpenCode Zen support with kimi-k2.5 and glm-5 for $10/month — powerful models without breaking the bank. And Ollama’s now an official provider. “Shipping faster than they can clone.”

🔌 Around the Web

AI News

📰 NVIDIA Launches Vera CPU, Purpose-Built for Agentic AI — NVIDIA’s first processor designed specifically for agentic AI and reinforcement learning — 2x efficiency and 50% faster than traditional CPUs. 88 custom Olympus cores, LPDDR5X memory at half the power, and a 256-CPU liquid-cooled rack that sustains 22,500 concurrent environments. Agents are getting their own silicon.
📰 Sam Altman says AI will eventually be sold like electricity and water — OpenAI CEO envisions a future where “intelligence is a utility” metered by token usage. Compute capacity becomes the gatekeeper of AI access, with major tech companies racing to build hundreds of billions in infrastructure to avoid pricing out the masses. Meanwhile, NVIDIA’s building hardware for your agents to run locally.

Essays & Analysis

✍️ Talk to Your AI: Realtime Voice Chat in TanStack AI — First-class support for realtime voice conversations with provider-agnostic architecture, client-side tool execution, and a React hook that makes building voice chat feel like building a form. OpenAI Realtime and ElevenLabs supported out of the box.

Tools & Resources

📖 Lazy Tool Discovery: Scaling AI Tool Systems Without Drowning in Tokens — TanStack AI ships lazy tool discovery: mark tools with lazy: true and the LLM discovers them on demand instead of receiving all definitions upfront. Saves tokens, reduces decision fatigue, and scales to dozens of tools without context window bloat.
🛠️ CanIRun.ai — Find out which AI models your machine can actually run. Browse 70+ models from Llama to DeepSeek with memory requirements, context windows, and hardware compatibility checks.
🛠️ Chrome DevTools MCP — Live Browser Session Support — Chrome DevTools MCP now connects to your real, signed-in browser sessions. No extensions, no headless browser, no separate logins — just one toggle. Agents can debug pages you’re already logged into, access selected elements from DevTools, and investigate network requests directly.

🌟 Community Highlights

The agent infrastructure stack keeps growing:

♟️ Claw Chess Cup — First Agent-Only Chess Tournament by @nicoletteduclar — Agents compete head-to-head in the first OpenClaw chess tournament. Register your agent at clawleague.live/skill.md and join the bracket. May the best model win.
🛠️ AuraWallet by @nicoletteduclar — Open-source secret manager for AI agents. Securely share passwords, API keys, and credit cards with agents — keys stay local, agents get scoped tokens (not raw credentials). Works with OpenClaw, Claude Code, Cursor, Codex, and any MCP-compatible agent.
🛠️ NemoClaw by NVIDIA — Open source stack for securely running OpenClaw always-on assistants. Sandboxed environment with Landlock + seccomp + netns, policy-enforced egress and filesystem access, and integrated NVIDIA Nemotron models via cloud, NIM, or vLLM.
🛠️ OpenClaw on DGX Spark — NVIDIA’s step-by-step playbook for running OpenClaw on DGX Spark (Grace Blackwell). Fully local agent with 128GB memory for large models — no data leaves your machine, no ongoing cloud API costs.

😂 Fun

My OpenClaw dragging me along

That’s your weekly claws! See you next week! 👋