Top AI Product

I’ve been keeping an eye on the whole OpenClaw security drama for weeks now — ever since Microsoft’s security blog dropped that warning about runtime risks back in February. So when [IronClaw](https://github.com/nearai/ironclaw) showed up on my radar, climbing to #3 on [trendshift.io](https://trendshift.io) and racking up 4.3k stars, I had to take a closer look.

Here’s the deal. IronClaw is built by NEAR AI — Illia Polosukhin’s team — and it’s essentially a ground-up rewrite of the OpenClaw concept, but in Rust, with security as the foundation rather than an afterthought. The core problem it addresses is pretty straightforward: OpenClaw exposes your credentials to the LLM context. That’s a fundamental architectural flaw, and no amount of patching can fix it. IronClaw takes a completely different approach. Every tool runs inside its own WebAssembly sandbox with capability-based permissions, and your credentials live in an encrypted vault inside a Trusted Execution Environment. The AI model literally never sees your raw API keys or tokens.

What I really appreciate is that it swapped out SQLite for PostgreSQL and went all-in on Rust’s memory safety guarantees. No buffer overflows, no memory corruption — entire categories of vulnerabilities just don’t exist here. It also has this neat dynamic tool building feature where you describe what you need and it generates a WASM tool on the fly.

The offline story is solid too. You can point it at a local Ollama instance and run completely air-gapped if you’re working with sensitive data. It also supports OpenRouter, Together AI, Anthropic, and basically any OpenAI-compatible endpoint.

It’s already getting traction on [Product Hunt](https://www.producthunt.com/products/ironclaw) and the broader developer community seems genuinely excited — not just about the security angle, but because it actually works well as an agent runtime. If you’ve been running OpenClaw and that Microsoft security warning gave you pause, IronClaw is worth a serious look. The install is dead simple: a one-liner curl script, cargo install, or just clone and build. Give the [repo](https://github.com/nearai/ironclaw) a spin and see for yourself.