Verified
Q1 Is the model endpoint registered and within its declared parameters?
Every AI inference becomes a governed, evidenced, replayable event — without replacing IAM, OAuth, your API gateway, GRC, or SIEM.
KYE™ adds a delegated-authority, evidence, and runtime-decision layer on top of the stack you already run.
2026-05-15 · v3.1 · KYE Protocol™ Ltd
From KYX to KYE™
KYC verifies the customer. KYB verifies the business. KYA verifies the agent. The natural next step was KYX — know your X. But KYX is still verification at onboarding. Once every component of a digital system — human, agent, model, dataset, tool, downstream API — can act under delegated authority, governance has to be a protocol, not a stitched-together report.
AI agents don’t just identify. They act. Existing systems verify who logged in. They cannot prove what was invoked, on whose behalf, against what dataset, under what policy, at what risk class, and whether the call is auditable. Read the full thesis →
Why existing tools are not enough
| Layer | Question it answers |
|---|---|
| IAM / OAuth | Can this identity access the system? |
| API Gateway | Can this request reach the service? |
| GRC | Are controls documented? |
| SIEM | What events happened? |
| MLOps platform | Did the deployment succeed? |
| KYE™ | Was this inference verified, allowed, against an approved dataset, in a permitted jurisdiction, under a compliant policy, at the right risk class, and is the call auditable? |
KYE™ does not replace existing infrastructure. It adds the missing per-inference accountability layer.
The KYE™ answer
The KYE™ Runtime Gateway™ resolves each one before the inference reaches downstream systems.
Q1 Is the model endpoint registered and within its declared parameters?
Q2 Is the caller authorised — under the delegated-authority chain — to invoke it?
Q3 Is the dataset being read or referenced approved for this purpose?
Q4 Is the tool, data, or downstream effect allowed in the caller’s jurisdiction?
Q5 Is the request compliant with the active Operating Model™ and policy bundle?
Q6 Is the model suitable for the declared risk class on its evaluation profile?
Q7 Is the entire call — inputs, outputs, decision, evidence — replayable from a signed Evidence Pack™? If any answer is “no”, the Runtime Gateway can deny, escalate, or shadow per the active mode.
The wedge
The Runtime Gateway™ runs in shadow mode — no production block. Every model-endpoint invocation is captured with full context.
Each observed inference becomes a signed Evidence Pack™: actor, principal, purpose, authority, scope, dataset, policy, decision — replayable.
Authority gaps, jurisdiction drift, and risk-class violations surface as events in your existing SIEM — without blocking.
Narrow guards deploy on the highest-risk model endpoints. Production stays fast; risk is bounded.
Runtime decisions block unauthorised, out-of-scope, or out-of-jurisdiction inferences before they reach downstream systems.
Roll out to additional workflows, sectors, and partners using the same protocol primitives and Evidence Pack™ format.
KYE™ ships in four Edge Governance modes (Shadow · Advisory · Guarded · Strict). You pick which.
First use case
Enterprise AI copilots Agentic payments Procurement agents Customer-support agents Compliance investigation agents Third-party AI workflows Open-finance workflows Clinical decision support
First commercial wedge: regulated AI workflows in financial services under EU AI Act, DORA, and the FCA Operational Resilience regime.
How it works
Entity hierarchy v3.1 (constitution §1a, locked 2026-05-15). The Runtime Gateway answers the seven questions against the leaf kye:endpoint:….
Tenant kye:tenant:…
Workspace kye:wsp:…
Model Endpoint kye:endpoint:… · policy_set + evaluation_profile + audit_stream
Runtime Gateway™ seven-question decision
Evidence Pack™ → Audit Stream kye:audstream:… → SIEM / GRC / Cold storage
A kye:model:… is the artefact. A kye:endpoint:… is the bound URL with its own governance — that is what gets verified.
The control plane
A Cloudflare-edge worker on the inference path. Resolves the seven Runtime Gateway questions in < 25 ms p95, with offline fallback.
Compiled Authority Bundle (signed at the edge) + Operating Model™ + Evaluation Profile + Policy Set for the resolved endpoint.
One signed Decision Map™ per call + a Risk Score (Ed25519) + an Evidence Pack™ entry pushed to the bound Audit Stream.
Shadow · Advisory · Guarded · Strict — per-endpoint, per-jurisdiction. Edge Arbiter enforces the Safety Floor when offline.
If a single question is unresolved at the edge, the Runtime Gateway never silently allows — it falls back to the declared mode and writes an Offline Evidence Log entry that reconciles on next sync.
Constitutional locks
Constitution §0.3 (adopted 2026-05-15). Every CI run, deploy, schema migration, secret rotation, and domain binding is governed by the same engines that govern customer inferences.
Governs every privileged action. Issues kye.purpose.request.v1, kye.purpose.admissibility.v1, kye.evidence.decision_map.v1, kye.replay.context_seal.v1. Same engine that authorises customer agents authorises our own deploys.
Every regulatory-framework claim maps to a control row + ≤ 90-day attestation. Emits kye.compliance.attestation.v1 and kye.evidence.pack.v1 on a fixed cadence. Frameworks: EU AI Act, NIST AI RMF, ISO 42001, DORA, SOC 2.
Customers do not have to trust our claims. They can replay them — against the same Evidence Pack™ format the Runtime Gateway emits for their own inferences.
The platform
Engines are the computation primitives. Rails are the user-facing surfaces that orchestrate them. Locked by constitution §14.
Entity · Authority · Purpose & Scope · State · Rules · Decision · Evidence · Replay · Directory · Ecosystem.
Purpose Permission™ · Resilience Loop™ · Directory · Operating Model™ · Delegated Auditability · Onboarding · Billing & Metering · Edge Governance™.
SDKs (TypeScript · Python · Go) · OpenAPI · MCP server · webhooks · Cloudflare-edge Runtime Gateway™.
KYE™ Cloud (app.kyeprotocol.com) · KYE™ Widgets (embeddable) · KYE™ Directory (search) · KYE™ Audit Pilot™ (regulator-facing).
Evidence Pack™ example
kye:prin:… that initiated the callkye:endpoint:… resolved at decision timekye.risk.score.v1kye.replay.context_seal.v1{
"evidence_pack_id": "kye:evp:acme.example:01JYEVP0...",
"issued_at": "2026-05-15T13:31:00Z",
"endpoint_id": "kye:endpoint:acme.example:claude-sonnet-eu-2",
"actor_principal": "kye:prin:acme.example:agent:ops-copilot",
"runtime_gateway": {
"mode": "strict",
"q1_verified": "true",
"q2_allowed": "true",
"q3_dataset": "approved",
"q4_jurisdiction": "eu-west",
"q5_policy": "compliant",
"q6_risk_class": "high-suitable",
"q7_auditable": "sealed"
},
"control_refs": [
"EU_AI_ACT.AIA_HIGH_RISK_LOGGING",
"EU_AI_ACT.AIA_HUMAN_OVERSIGHT",
"DORA.ICT_RISK_REGISTER",
"ISO_42001.AIMS_8_2"
],
"risk_score": {
"value": 12,
"out_of": 100,
"signature_suite": "ed25519-2025"
},
"replay_context_seal": "sha256:9c8a..."
}
Every Evidence Pack™ is replayable offline — auditors verify against the canonical format, not your screenshots.
Advanced assurance
Simulate what could happen under proposed authority changes before they ship to production.
Brief approvers with full context, alternatives, and risks before the inference executes.
Detect when a locally valid system has decoupled from current ground truth.
Surface slow, plausible drift that single-action audits miss — rolling-window detector.
KYE™ can simulate what could happen, brief approvers before action, and detect when a locally valid system has become decoupled from current reality.
Enterprise value
AI use-case inventory
Risk register
Control mapping
Audit evidence
Framework readiness (EU AI Act, DORA, ISO 42001)
Per-inference visibility
Runtime authority logs
Revocation cascade
SIEM events
Risk-score alerts
SDKs (TS / Py / Go)
OpenAPI
MCP server
Webhooks
Evidence Pack™ automation
Governance coverage %
Audit readiness
Risk-class exposure
Reduced AI adoption friction
Replayable claims
Pilot offerings
Every pilot has its own onboarding agent, demo seed pack, dashboard preset, Evidence Pack™ sample, and signed final report. The commercial menu is sent to qualified applicants by email — not published.
| Offering | Duration | Best for |
|---|---|---|
| KYE™ Discovery Workshop™ | 1 week | Use-case shortlist, authority-risk map, pilot scope. |
| KYE™ AI Agent Audit Pilot™ | 4 weeks | Enterprise AI / copilot / agent owners — one workflow, Shadow mode. |
| KYE™ Regulated Workflow Pilot™ | 6 weeks | Banks / insurers / healthcare / public sector — 2–3 workflows. |
| KYE™ Governance Ops Pilot™ | 8 weeks | GRC / security / legal / audit teams — inventory + register + dashboard. |
| KYE™ Edge Governance Pilot™ | 8 weeks | Defence / energy / drones / critical infrastructure — offline / degraded. |
| KYE™ Strategic Sandbox Pilot™ | 12 weeks | Banks / ecosystems / regulators / national platforms — multi-team. |
Pilot fees credit 100% against the first-year annual licence when signed within 60 days — the commercial menu and conversion terms are shared after qualification.
Call to action
After you apply, the qualification team responds within 2 business days with the full commercial menu, conversion terms, and pilot scope draft.
Start with evidence. Move to enforcement when ready.
Canonical KYE™ surfaces referenced on this page: Audit Pilot · Delegated Auditability · Evidence Pack™ · KYE™ Cloud™ · KYE™ Directory™ · KYE™ Edge Governance™ · KYE™ MCP Server · KYE™ Onboarding Agent · KYE Protocol™ · Purpose Permission · Resilience Loop.