REMAP wraps the agent you already have and moves its decisions onto a deterministic plane — no tokens, no GPU, every decision audited. The model keeps writing. It stops deciding.
Sign up holds your place in line · AIRUT — the concierge on this page — runs on MDIL and answers anything about REMAP, live
Sign up — first 1,000 places.
In order of arrival. One confirmation email, then silence until your place comes up.
MDIL Labs Oy (Helsinki) stores these details to manage the REMAP waitlist and contact you about your place — nothing else. Ask info@mdil.ai to update or delete them at any time.
{{ confirmCopy }}
Reasoning separated from generation.
The decision is made before the LLM is invoked. Routing, guards, privacy, are-we-done — deterministic computation, no model in the loop. A model is called only for the part that genuinely needs writing.
None, on the common path. The decisions run on a single CPU core, in under a gigabyte — tens of instances side by side, near-real-time. It runs on the hardware you already own.
Decisions are made by declared, deterministic detector ensembles — MMTs — not trained weights. Every result lands in a hash-linked, tamper-evident audit chain built for regulator review.
REMAP wraps the exact points where an agent goes off the rails and moves those decisions onto the deterministic plane. Your framework, tools, prompts and code stay.
Zero-Trust reasoning infrastructure for agentic AI.
REMAP is the first product built on MDIL — a deterministic reasoning layer any agent stack can run on. Zero-trust operation is its flagship profile.
No model in the decision loop — authored logic that can be read, replayed and verified against what it declares.
What did it decide. Why. Show me. Hash-linked and tamper-evident — the three regulator questions, made boring.
Sensitive data is pseudonymized before it reaches any model — zero entity bytes cross the line.
Cloud, on-premise, or fully air-gapped — identical behavior. Sovereignty as the deployment diagram, not a slogan.
Ask the thing itself.
Ask Airut anything about REMAP or MDIL — what it does, how the audit chain and Symbol Security work, or how to claim a place.
A decision plane with nothing to take on faith.
No learned components on the plane — deterministic by construction, and what that buys under adversarial pressure.
The adversarial-technique matrix, mapped — which techniques a deterministic plane has the potential to mitigate, one by one.
Pseudonymization before the model — threat model, canary probes, and the recall benchmark.
Hash-linked, byte-reproducible replay — how every decision shows its work, and how an auditor checks it.
The white paper and chapter briefs open here, and the full PDF lands with your confirmation — no separate request.
Limited places. In order of arrival.
Helsinki, Finland
Zero Trust White Paper ↑