I build runtime infrastructure for AI agents — the layer between the model and the chaos.
Right now I'm working on Identity OS, a behavioral middleware that gives AI agents persistent personality, drift detection, and stress-adaptive guardrails. It sits on top of LangGraph, CrewAI, or OpenAI Agents SDK and tells your agent what it can and can't do — every single turn.
Your agent sends behavioral observations. Identity OS returns a read-only ExecutionContract — a constraint set covering allowed actions, decision style, energy, stress state, and drift status. Deterministic. Auditable. Framework-agnostic.
In adversarial testing, Identity OS blocked 100% of forbidden actions across multiple agent frameworks. Without it, 100% executed unchecked.
| 1,200 benchmark turns | 4 adversarial scenarios, all assertions pass |
| 100% enforcement | Every forbidden action blocked in E2E testing |
| Multi-level drift detection | Catches behavioral deviation in real time |
| Adaptive stress model | Automatically tightens constraints under pressure |
| Multiple persona archetypes | Rich personality system out of the box |
Python · TypeScript · FastAPI · PostgreSQL · LangGraph · CrewAI · Docker
- Email — xiocasso@outlook.com
Building the guardrails so agents can be themselves — safely.