We're building the safety layer AI agents never got.
LLMs got smarter. Agents got autonomous. Nobody built the brakes. So we did.
Our mission.
AI agents are shipping to production faster than anyone predicted. They book meetings, write code, move money, and call APIs — on their own. And right now, most of them run with zero spending controls, no kill switches, and no human approval before high-stakes actions. We started Wickd because we watched a single agent burn through $47,000 in a weekend, and realized the tooling to prevent it simply didn't exist. Gateways protect infrastructure. Firewalls protect networks. Nothing protects the agent itself. Wickd is the runtime safety net — budget limits, kill switches, and approval gates that live inside your agent code. No extra infrastructure. No latency. Just a pip install and a decorator between you and a very expensive mistake.
How Wickd protects your agents.
One SDK. Three layers of defense. Embedded directly in your agent - not bolted on from the outside.
Small team. Big opinion about agent safety.
We're engineers, researchers, and builders who've shipped AI in production — and cleaned up the mess when it went wrong. We obsess over reliability, write proofs for our budget algorithms, and believe open source is the only way to build trust in safety tooling.