Control what AI makes real.

The direct state change this action would create now. Limren checks whether that immediate effect is allowed before the change can reach reality.

The capability, access, or authority this action creates for later actions. Limren treats newly created power as part of the decision, not as an invisible side effect.

The combined result of this action with other pending or committed changes. Actions that look safe individually can become unsafe together, so Limren evaluates the combined effect.

How strongly the target can guarantee the final precondition at write time. STRICT uses target-enforced preconditions; FENCED relies on exclusive control; NON-STRICT cannot guarantee atomic enforcement.

Whether the actor still has valid permission to make this exact change at the moment of commit. Earlier approval does not override expired or changed authority.

The exact real resource, record, system, or object the action intends to change. Limren binds the proposal to that target so the commit cannot silently drift somewhere else.

The exact mutation being proposed: operation, values, and intended effect. Limren binds the committed write to the payload that was actually reviewed and validated.

The external or internal facts this change depends on. If a dependency has changed since the proposal was evaluated, Limren stops and rechecks before commit.

Whether the decision is still true at the last possible moment. Limren rechecks time, authority, target state, dependencies, and bindings immediately before reality changes.

A controlled flow · Five stages, one continuous system

One action.
A chain of consequences.

Explore what could happen, understand what else changes, and check what is safe before anything becomes real.

BEFORE AN AI CHANGES REALITY

01INTENT

A proposed actionThe starting point

02EFFECTS

Possible outcomesDirect + indirect

03DEPENDENCIES

What else changesConnected state

04DECISION

Last possible checkPolicy · drift · authority

05REALITY

Observed outcomeCommit, then verify

POSSIBLE CONSEQUENCES VIOLATION

A failed invariant interrupts the proposed action before commit. No real-world change occurs.

The graph is a visual illustration of the decision process.

The commit moment

Nothing becomes real until COMMIT.

The next state can be proposed. The real state does not change until Limren checks it at the last possible moment.

Reality answers back

Reality answers back.

Limren compares what was committed with the state that actually returned from the real system.

Universal control

Any intelligence.
One commit boundary.

Agents, workflows, APIs, humans, systems and devices can propose. They cross the same boundary before reality changes.

INTENT

REALITY

CURRENT STATE

Δ STATE

PROPOSED STATE

AUTHORITY STATE DEPENDENCIES VALIDITY

LAST POSSIBLE CHECK

COMMITTED

INTENDED STATE

OBSERVED STATE

VERIFIED

MATCH → VERIFIED · DIFFERENCE → DRIFT

CAN PROPOSE

CAN BECOME REAL

AGENT

WORKFLOW

API

HUMAN

SYSTEM

DEVICE

REAL STATE

ANY INTELLIGENCE CAN PROPOSE. LIMREN CONTROLS WHAT BECOMES REAL.

Autonomy,
with a boundary.

Architecture GitHub

Control what AI makes real.

One action.A chain of consequences.

Nothing becomes real until COMMIT.

Reality answers back.

Any intelligence.One commit boundary.

Autonomy,with a boundary.

One action.
A chain of consequences.

Any intelligence.
One commit boundary.

Autonomy,
with a boundary.