MergeOn
Free Trial
Substrate Overview

How the MergeOn Substrate Works.

Every claim the substrate makes traces to evidence. Every dependency between facts is tracked. Every operation is replay-safe. The substrate is what makes enterprise AI safe to act inside.

SOC 2 Aligned
HIPAA Ready
Bring Your Own Storage
Evidence Chains
The four principles

What the substrate guarantees

The substrate is defined by what it makes provable, not by what it lets a model improvise.

Truth lives outside the model

Operational knowledge is stored in the substrate, not in model weights. Reasoning engines read from it; they never own it. Provider changes do not invalidate the record.

Outputs are evidence-linked

Every value produced by the substrate resolves to source coordinates and a verifiable hash. No claim reaches an operational system without provenance.

Dependencies are first-class

Cross-document references are tracked as a graph. A change in one fact propagates to every downstream operation that depended on it.

Operations are replay-safe

Every decision is logged with the exact context that produced it. Regulators, counterparties, and auditors can reconstruct any answer.

The category-defining sentence
Truth lives outside the model.
Semantic normalization

Canonical, typed, version-aware

Before reasoning happens, the substrate resolves operational concepts into a canonical, versioned typing. The same engines serve every vertical; only rule packs differ.

Canonical operational identity

A single canonical record per party, asset, and obligation — across systems, divisions, and acquisitions.

Coordinate-anchored extraction

Every field traces to exact source coordinates, retained for replay and verification.

Version-aware semantics

Amendments and supersession chains resolve to the right effective version at every point in time.

Canonical type system

Concepts are typed and bound to rule packs, so verticals share engines and differ only in policy.

Evidence

Every output points at the pixel

Every value produced by the substrate resolves to source coordinates and a verifiable hash. No operational claim escapes without provenance.

Source Document
Page 3, coords [142, 287, 445, 312]
Extracted Field
purchase_price: $1,250,000
Verifiable Hash
e3b0c44298fc1c149afbf4c8…
Operational Output
The substrate exposes the value to downstream operations.
Linked back to source and hash at every step.
If you cannot point to the pixel, you cannot claim the fact.
The substrate refuses to surface answers that lack provenance. Auditors and regulators read a record, not a guess.
Dependency intelligence

A change in one fact propagates to every fact that depends on it

Cross-document references are tracked as a typed graph. The substrate computes the impact of every change before it reaches downstream systems.

Purchase Agreement
closing_date
March 15, 2026
Loan Commitment
funding_deadline
closing_date − 3 days
← derived from Purchase Agreement
Title Insurance
effective_date
closing_date
← derived from Purchase Agreement
If closing_date changes, the substrate propagates the change to every dependent fact, flags affected counterparties, and records the cascade for replay.
Governed execution

The substrate doesn’t just observe — it constrains action

Reading is governed by policy. Writing is governed by policy. Every model-initiated tool call is mediated before it reaches a system of record.

Policy-mediated tool calls

Every model-invoked action is evaluated against policy before it reaches a system of record.

Deterministic orchestration

Operational workflows execute against rule packs, not against model improvisation.

Governed AI memory

Memory is versioned, scoped, and bounded by policy. Never fused into model weights.

Replay-safe events

Every event is reconstructable from logged context, inputs, and policy decisions.

Substrate in Production

AI-LTOR — MergeOn’s First Real Estate Vertical Operating System

AI-LTOR is MergeOn’s first real estate vertical operating system, built on the MergeOn substrate. It exists to prove the architecture under the conditions of an actual regulated industry.

Governed communication substrate

Every message between parties is policy-mediated and evidence-linked.

Relationship intelligence

The substrate tracks who depends on whom across the lifecycle of a transaction.

Replay-safe operational events

Every operation is reconstructable from logged context and inputs.

Governed AI memory

Memory is versioned, scoped, and bounded by policy — not fused into weights.

Deterministic orchestration

Workflows execute against rule packs, not against model improvisation.

Canonical operational identity

A single canonical record per party persists across systems and acquisitions.

Multi-engine coordination

Reasoning, retrieval, and policy engines coordinate through the substrate, not through any single model.

Positioning note
AI-LTOR is not MergeOn. MergeOn is the parent substrate. AI-LTOR is MergeOn’s first real estate vertical operating system built on the MergeOn substrate — proof that the architecture holds under the conditions of an actual regulated industry. Additional vertical operating systems may follow in other industries.

See the substrate in action

These are not slides. Request a technical walkthrough and we will show you the substrate operating against real corpora.

Request a walkthroughRead the reference architecture