Architecture

De-risk. Govern.
Attest.

AI proposes a response or action. SASKI applies the organization’s rulebook, returns an enforceable decision, and records why it was made.

The shared architecture

One control model from proposal to proof.

01 / OBSERVE

Request or proposed action

Capture the interaction, context, and intended operation at the application boundary.

02 / GOVERN

Versioned rulebook

Evaluate authority, scope, parameters, prerequisites, privacy, and configured safety rules.

03 / DECIDE

Allow, hold, or deny

Return a deterministic decision the application can enforce before continuing.

04 / ATTEST

Inspectable evidence

Record the rulebook version, checks, outcome, reason codes, and approval state.

REQUESTRULEBOOKDECISIONEXECUTION REFERENCEATTESTATION

The rulebook

Turn organizational boundaries into explicit checks.

The model can interpret context and propose a path. Authority comes from a separate, versioned rulebook.

Identity & authority
Who is requesting the interaction or action, and what are they permitted to do?
Context & data
What information, user context, jurisdiction, and configured protections apply?
Scope & parameters
Is the target, operation, amount, destination, or response within allowed bounds?
Prerequisites & holds
Which conditions or human approvals must be satisfied before the workflow continues?
Output & evidence
What may be returned, what was actually decided, and which records must be retained?

Three deployment paths

Use the same governance model at different stages.

01 / Live interactions

SASKI SDK

Govern input before the model and validate output before the application returns it to a person.

Explore SASKI SDK
02 / Live actions

SASKI Agentic

Scan input and output, and enforce the rulebook before an agent reaches a connected tool or system.

Explore SASKI Agentic
03 / Evaluation

SASKI Replay

Run historical records through a selected rulebook to inspect decisions before live enforcement.

Explore SASKI Replay

Evaluate before enforcement

See where the rulebook would intervene.

Replay and Shadow Mode help teams examine representative traffic without changing the live response or action path.

Review detected risk signals, proposed interventions, reason codes, and sample attestations. Use the results to refine the rulebook, define escalation ownership, and plan the integration boundary.

Evaluation evidence describes the records and configuration that were tested. Production coverage begins when the relevant live paths are routed through the control layer.

Explore SASKI Replay ↗

Decision and execution

Connect policy evidence to system evidence.

Illustrative decision record

Request
PO-250K / new vendor
Rulebook
Enterprise purchasing v4.2
Decision
HOLD
Reason
Vendor status + approval threshold
Approval
Executive review required
Execution
Not yet attempted

A governance receipt proves the decision the control layer made. The application’s execution record establishes what happened next. Keeping both prevents a generated claim from being mistaken for a completed action.

Implementation matters

Make the governed path the real path.

Define the boundary

Identify protected interactions, actions, data flows, authority, failure behavior, and approval owners. Account for alternate routes that could bypass the control layer.

Verify the path

Test allowed, denied, held, and failure scenarios. Correlate SASKI attestations with application and execution logs so reviewers can distinguish a governance decision from a completed action.

One architecture, four products

Apply the model where it matters.

01 / Agent governance

SASKI Agentic

De-risk agentic applications by enforcing rulebooks across input, proposed actions, and output—with attestation for each governance decision.

Explore SASKI Agentic
02 / Interaction governance

SASKI SDK

De-risk human-facing AI with rulebooks before and after the model, plus attestation for the resulting decisions.

Explore SASKI SDK
03 / Evaluation & attestation

SASKI Replay

Test historical interactions and actions against rulebooks, then inspect the resulting governance evidence.

Explore SASKI Replay
04 / Vertical application

SASKI Estate

Apply the same rulebook governance and attestation architecture to AI-enabled smart homes.

Explore SASKI Estate

Start with your use case

What should your AI
be allowed to do?

Let’s talk