An AI can ask.
The kernel decides whether it has the authority to act.
Trunomy OS enforces kernel-level authorization and deterministic validation for autonomous AI agents, A proposal to solve the problem of rogue AI agent alignment..
STAGE
Phase Pre-Seed
ARCHITECTURE
Microkernel Engine
SECURITY
Capability-Based OS
FOUNDER
Roberto C. Aleman F.
AI Agents are Gaining Autonomy, But Lack Kernel Protection
Current agent frameworks rely on soft application-layer prompt guidelines or trust LLM function calling directly with API keys and root credentials.
Prompt Injections
LLM outputs cannot be treated as security decisions. Adversarial context hijacking easily overrides system instructions.
Over-Privileged Execution
Agents are given direct broad access to DBs, filesystem paths, and APIs without fine-grained handle checks.
Unchecked Lateral Movement
Once compromised, an agent can cross multi-tenant boundaries, escalate privileges, and execute unsafe system modifications.
Intent ≠ Authority. The Deterministic Microkernel
Makes a Request
(Non-deterministic Intent)
- Operation: read / write / exec
- Target Resource: data, service, tenant
- Handle ID: 0x8F3A2...
- Context: gen=4, tenant_id=B
- Validate ABI & Pointers
- Resolve Capability / Handle
- Check Generation & Tenant
- Enforce Quotas & Revocation
Why Trunomy Cannot Be Replaced by Application Wrappers
Trunomy OS operates it operates at layer 0, inserting deterministic capability handles beneath the execution layer.
ABI & Memory Isolation
Intercepts system calls at the Application Binary Interface (ABI). Pointers and memory ranges are verified before CPU dispatch.
Capability Handles
Agents do not hold long-lived tokens. They hold ephemeral, unforgeable capability handles tied to generational counters.
Tenant Generation Tracking
Cryptographic verification ensures an agent's execution context never leaks into neighboring multi-tenant workloads.
Instant Quotas & Revocation
Kill switches operate in microseconds at the kernel level without relying on the LLM or orchestration server response.
Built For Enterprise AI Infrastructure
Unlocking deployment in mission-critical industries where uncontained AI error is unacceptable.
AI Agent Security
Autonomous agent frameworks requiring strict runtime sandbox boundaries.
Multi-Tenant Cloud
SaaS platforms hosting agentic workflows across thousands of untrusted tenants.
Sensitive Data Access
Fintech, Healthcare RAG, and Legal AI querying protected data stores.
Controlled Ops
DevOps, autonomous coding assistants, and automated cloud infrastructure agents.
CATEGORY OPPORTUNITY
Trunomy is the first operating system designed to protect the host from rogue AI agents and prevent data leaks or major risks; consequently, its primary strength lies in the very feature that is most in demand today and will be in the near future...
Trunomy OS In Action
Demonstrating real-time microkernel capability enforcement, memory bounds protection, and immediate syscall interception.
Building the Kernel Layer for the Agentic Era
Led by Roberto C. Aleman F., a Computer Engineer with 20+ years of deep technical expertise in systems engineering, low-level architecture, cryptography, and systems programming.
Roberto C. Aleman F.
Founder & Chief Architect
"AI needs more than application wrappers—it requires deterministic, hardware-enforced boundaries at the operating system level."
Join Us Early
We are raising our initial Pre-Seed capital round to expand low-level microkernel development, launch developer SDKs, and run enterprise pilot programs.