Engineering reliable systems for AI agents.
I build RCOS (Recursive Capability Operating System): an evolving execution and verification system for agentic work. The focus is what happens after an agent proposes an action—authority, routing, durable execution, independent evaluation, and inspectable evidence.
The design separates the agent from the system that authorizes, executes, and verifies its work. It aims to remain portable across models, agent interfaces, and compute environments.
Governed execution for agents and coding workflows.
Agent → Authority & routing → Execution → Evaluation → Evidence
Public engineering work
- DSH Operator UI — RCOS operator integration. The public package is reproducible against its documented DeepSeek Harness web-profile version; native Desktop integration remains a development preview.
- FlowRouter Capability Manifest — Versioned specification, JSON Schema, examples, and validation tooling for portable capability contracts and their evidence.
Release status: The canonical RCOS source is private pending publication, security, licensing, and fresh-install verification. The public repositories above represent specific components, not a general-availability RCOS release.
Agent infrastructure · software factories · evaluation systems · local inference