ES
Trajectory enforcement for AI agents

Govern the trajectory.
Not just the next tool call.

DiaCroma evaluates every action routed through it against what the agent — and the agents around it — have already done. It adds durable, deterministic enforcement without replacing your agent platform, identity system, runtime or cloud.

Per-action controls ask, “Is this call allowed?” DiaCroma also asks, “Is it still allowed after everything already done?”

Interposed on governed routesDecides before executionSurvives restartsReconciles the real effectRuns in your network

The control loop

Every action routed through DiaCroma passes through five explicit states.

The agent still reasons in its own runtime. DiaCroma governs the transition from proposed intent to business effect.

01 · DECLARE

Sign the authority.

Mission, tools, caps, budgets, resources, invariants and acceptable evidence become a versioned anchor.

02 · PROPOSE

The agent asks to act.

A tool call, delegation, message or other effect arrives at a boundary outside the model.

03 · EVALUATE

Read the live trajectory.

DiaCroma checks cumulative authority, sequence, lineage, shared constraints and evidence against durable state.

04 · DECIDE

Return a deterministic verdict.

The structural decision does not depend on a language model.

ALLOWWARN_REPLANBLOCK_ESCALATE
05 · SETTLE

Close against the real effect.

The permit is consumed, the resulting effect is reconciled, and the evidence row is sealed for replay.

A permit is valid only against the live trajectory state that produced it. If that state moves first, the proposed effect is no longer authorized.

Product boundary

What your platforms keep.
What DiaCroma adds.

DiaCroma consumes identity, policy and tool context from the surrounding stack. It does not rebuild the layers enterprises already operate.

YOUR CONTROL PLANE

Identity and lifecycle

  • Agent identity and ownership
  • Registry and inventory
  • Permissions and Conditional Access
  • Runtime and orchestration
YOUR PER-ACTION CONTROLS

One proposed call

  • Tool allowlists and argument caps
  • Gateway and MCP policy
  • Content and data controls
  • Sandbox and supply-chain controls
DIACROMA · TRAJECTORY ENFORCEMENT

The cumulative system

  • Authority spent across actions and agents
  • Sequence, delegation and shared budgets
  • Cross-agent and cross-system lineage
  • Effect settlement and deterministic replay

Where it runs

Local boundaries.
One signed trajectory.

Agents and business tools stay where they run. DiaCroma places an execution boundary near the effecting route while the same signed authority, trajectory state and evidence follow the work across platforms.

EXISTING ESTATE

Agent runtime

A runtime connected through a validated adapter proposes an action. Platform validation is reported separately.

→
LOCAL ENFORCEMENT

DiaCroma boundary

Builds context, reads the live trajectory and returns a permit, replan or block.

→
BUSINESS EFFECT

Tool or system

On a closed governed route, executes only with a valid permit and returns the receipt required for settlement.

Customer-controlled placementBoundary and state can run in infrastructure you operate.
Shared evidenceEvery decision and effect joins the same recomputable trajectory record.

It runs in your network, at your boundary; validation is reported route by route. See what is running today, and what is coming next, on the Evidence page → · What changes in your architecture →

What the enterprise signs

What you do not declare
is not governed.

REQUIRED

Authority and identity

Mission anchor, deployer credential, agent identity, allowed tools and the version of the authority being enforced.

LIMITS

Caps and cumulative budgets

Exact per-argument ceilings plus the authority that may be spent across the lifetime of a trajectory or delegation tree.

RELATIONSHIPS

Resources, lineage and sequence

What agents read and write, which resources are coupled, and which operation classes cannot follow one another.

EVIDENCE

Claims, receipts and settlement

What each tool can prove, how fresh that evidence must be, and how the proposed action closes against the effect that actually occurred.

Frequently asked, answered precisely

What DiaCroma is —
and what it is not.

What category is DiaCroma creating?

Trajectory enforcement for AI agent systems: stateful, pre-execution control over accumulated authority, sequence, lineage, delegation and shared constraints.

Is DiaCroma an agent gateway?

It can use a gateway as an enforcement point, but the gateway is not the product. DiaCroma maintains the signed authority, trajectory state and evidence the boundary enforces.

Does it replace agent platforms or control planes?

No. Identity, inventory, runtime, orchestration and cloud administration remain where they are. DiaCroma adds a shared trajectory layer across governed routes.

Is it only a mission-drift monitor?

No. Drift is one case. DiaCroma also enforces cumulative budgets, sequence, evidence, delegation and cross-agent constraints before execution.

Does a language model decide whether an action is blocked?

Not for structural controls. Tool membership, numeric caps, cumulative budgets and sequence are evaluated deterministically. Optional semantic signals are identified separately. A model reads and proposes; a person signs. At runtime, the decision is deterministic and can be recomputed.

Next

Walk the six screens
with one agent.