Obsta Labs builds kernel-level security and control primitives for AI agents. It is at the founder-led stage: one person building the product and technical direction, while talking to early collaborators, advisors, and future team members.
Sole founder, building kernel-level security and control primitives for AI agents. The flagship is Bulwark — an open-source read-consent gate that pauses a protected file open at the Linux/macOS kernel before any byte reaches an agent.
Bulwark did not appear in isolation. It was built on a working stack I run every day: NeuroRouter Pro (context for live model sessions) and Hiveram (the shared work graph and provenance) drove the build itself, and VectorCourt served as the adversarial review gate throughout. Hivebus (open source) is the coordination fabric underneath. These tools are mature because I depend on them — Bulwark is what the stack produced.
Obsta Labs Pte. Ltd. — Singapore (UEN 202622479E).
Also operating as Obsta Labs LLC — Delaware, USA.
Web: obstalabs.dev · Contact: hello@obstalabs.dev
Current focus includes Bulwark, an open-source read-consent gate for AI agents; VectorCourt, a decision governance engine; and CoreGate (obsta.ai), a control plane for regulated agentic banking.
Building something at the intersection of AI agents and systems security? Obsta Labs is talking to early collaborators and advisors — hello@obstalabs.dev.