Governed Event Execution

Control high-risk decisions before they become business risk.

StreamKernel governs the boundary where an AI or operational decision becomes an action. It enforces policy and authority before execution, then binds evidence to what actually happened.

Find Your Use Case See How It Works

Keep the systems you already use. Govern the transition between recommendation and action.

RecommendationConsequential action

Governed Execution Boundary

PolicyAuthorityEvidence

Governed execution

Find the boundary

Where does a recommendation become a consequential action?

That is where StreamKernel fits. The technologies around the boundary can change. The boundary does not.

Transport- and vendor-agnostic. Kafka is one integration pattern, not a requirement. Connect through supported or purpose-built adapters and executors.

See the customer architecture

Upstream systems Fraud engine · Agent · Model · Rules engine · Application Evaluate / recommend

StreamKernel Policy → Authority → Control → Execution → Evidence Govern before action

Downstream systems API · Payment system · Case platform · Database · Enterprise app Act / commit

Start here

What problem are you trying to solve?

Start with the operational risk, not the platform.

Recommended starting point

Bind policy version, model identity, transform chain, provenance, and disposition to the execution itself, so the decision is reconstructable from one record.

See the closest environment

Choose your path

Choose the environment closest to yours

See how the same governed execution pattern maps to your operating reality.

Financial Services

Govern and prove high-risk decisions before downstream action.

Put enforceable policy around your existing fraud, AML, sanctions, credit, and risk engines
Bind decision evidence to the execution instead of reconstructing it from logs
Keep regulated data and inference inside your own boundary

Explore Financial Services

Decision request

Policy + evidence

Downstream action

99,102 records/secGoverned data path · fraud/AML screening · Apple M5 Pro · source-capped 3-run mean · CV 0.827% 1 receiptGoverned action evidence · policy, authority, decision, route, model, provenance, and disposition bound to one execution 25,488Governed agent tool-audit events in a completed proof run 1 JVMCompact governed runtime · no required SaaS control plane

Two proof families, kept distinct: governed data movement measures enrichment, policy, provenance, and delivery; governed consequential action measures authority-bound execution against a protected system. Every published number carries its hardware, date, run count, semantics, and measurement boundary on the benchmarks page.

Execution flow

One governed path. Evidence on every event.

Controls execute in the event path—before a protected system receives the result.

01
Ingest
Events enter from an approved source.
02
Decide
Policy, provenance, and cost are evaluated.
03
Enforce
Actions are applied before downstream delivery.
04
Deliver
Only governed results continue to protected systems.

Governed Event ReceiptSynthetic sample

Decision: Decline
Route: AML case review
Evidence: Bound to execution

See what the receipt records

The artifact

If this decision is challenged later, can you prove what happened?

A Governed Event Receipt is the record StreamKernel preserves for one governed execution. It is written while the decision executes, not reconstructed afterwards from logs, traces, and sink records.

Event identityEvent id, pipeline id, run id, timestamp, source and sink
PolicyPolicy version, policy SHA-256, decision, route, case priority, reason codes
Model contextModel name and version, runtime, in-process execution, model reference digest
AuthorityEnforcement model, fail-closed posture, authorization granularity, pipeline identity
TransformThe transform chain that actually ran on the event
Payload evidenceSource-text digest and redaction posture rather than the sensitive payload itself
ProvenanceThe streamkernel.provenance.* headers emitted with the governed event
Audit chainHash algorithm, canonicalization, previous and current event hash, audit path
DispositionWhere the event went downstream, and whether it was routed to DLQ

governed-event-receipt.financial-services.sample.json Synthetic

"final_decision":     "DECLINE",
"decision_route":     "aml_case_review",
"case_priority":      "P1",
"policy_version":     "financial-credit-fraud-aml-v1",
"policy_sha256":      "sha256:3ab41c9d...",
"aml.sanctions.hit":  true,
"transform_chain":    ["DJL_EMBEDDING",
                       "EMBEDDING_TO_WIREEVENT",
                       "CREDIT_FRAUD_AML_DECISION"],
"enforcement_model":  "inline_fail_closed",
"event_hash":         "sha256:6d2a91fb..."

Inspect the full receipt Download JSON

Sample data is synthetic. Field names come from the credit/fraud/AML decision-event pipeline in the StreamKernel runtime. StreamKernel does not produce the fraud, AML, or credit score itself — it governs the execution around whatever engine does, and preserves the evidence.

Latest guidance