Delentia OS:
Deterministic AI
Guardrails
for Agentic Systems
A live MCP gateway enforcing mathematical authorization (FDIA), structured reasoning (RCT-7), and policy control for AI agents — installable via npx today. Part of a broader constitutional AI architecture still in active development.
Interactive MCP Kernel & FDIA Safety Gate
A browser-only simulation: a few keyword rules and a fixed data score. Nothing is sent to a server and nothing is signed. It only shows how the equation F = D^I * A behaves.
What you can use now, and what does not exist yet
Delentia is at an early stage, built by one person. It has no enterprise customers and no hosted service. We say so plainly so you can decide on facts.
The agent runtime, its safety gate and a signed audit trail. Runs on your own machine.
- pip install delentia-os
- delentia demo run compare --scripted (no model, no key)
- Human-signed approvals before risky actions
- Hash-chained, signed audit log you can verify yourself
- Works with a local model or your own provider key
We set it up with you on one real workflow, measure it against a plain agent, and publish the result only with your permission.
- One workflow, run end to end
- Pass rule written before the first run
- Results reported per workflow, failures included
- Your data stays under your own policy
There is no hosted Delentia service and no SLA today. Leave your address and we will tell you if that changes.
- No credits, plans or billing exist yet
- No uptime promise is made
- No compliance certificate is claimed
Every number comes with its method and its limits
No result with a capable model exists yet. What we have comes from scripted models and small local models, and each result says so.
Run it in two minutes
pip install delentia-os, then delentia demo run compare --scripted. No model and no key needed; the demo shows the safety gate, a human signature and a signed audit log working for real.
