Deterministic governance for AI outputs — with audit-grade reasons.

Insight Guard is a governance kernel you call at runtime to decide whether to allow, block, cooldown, or no-op. It is built for contract compliance: stable reason codes, fixed behavior versions, tenant isolation, and kill-switch readiness — without storing raw user text by default.

Docs = Contract

Deterministic behavior_version

Stable reason_code enum

Tenant isolation & minimal data

Fail-safe & graceful degradation

Kill switch (off / shadow-only)

Request access / reference pack View contract snippet View pricing Read notes

Why Insight Guard

Most “safety layers” drift silently. Insight Guard treats governance as a contract: behavior is versioned, reasons are enumerable, and audits are replayable.

What it controls

High-risk generations, repetitive “insight” injections, compliance-sensitive outputs, and any decision that must be explainable to security, legal, or procurement.

What it refuses to be

Not a platform. Not a dashboard. Not “magic safety.” No hidden policy changes. No consulting disguised as product.

How it works

Single endpoint: your app calls Guard with context and optional hashes.
One of four outcomes: allow / block / cooldown / no_op.
Audit UX for machines: stable reason_code, audit_id, and behavior_version.
Fail-safe: configurable fail-open / fail-closed for timeouts or errors.
Minimal data surface: default is “no raw text storage”. Store hashes / metadata only.
Kill switch: tenant-level off or shadow-only (log without enforcement).

{
  "decision": "allow | block | cooldown | no_op",
  "reason_code": "STABLE_ENUM",
  "audit_id": "aud_...",
  "behavior_version": "v1.0.0",
  "latency_ms": 12,
  "policy_mode": "enforce | shadow | off"
}

Contract snippet (minimal)

POST /v1/guard
Content-Type: application/json
Authorization: Bearer <API_KEY>

{
  "tenant_id": "tnt_123",
  "event": {
    "type": "assistant_output",
    "content_hash": "sha256:...",
    "context": {
      "channel": "prod",
      "model": "vendor/your-model",
      "route": "reply",
      "user_tier": "free | paid"
    }
  },
  "options": {
    "fail_mode": "open | closed",
    "shadow_only": false
  }
}

200 OK

{
  "decision": "cooldown",
  "reason_code": "COOLDOWN_ACTIVE",
  "audit_id": "aud_6f3c...a91",
  "behavior_version": "guard.v1.contract",
  "policy_mode": "enforce",
  "expires_in_sec": 3600
}

What we guarantee

Stable semantics: old behavior versions remain valid; no silent drift.
Reason codes do not rename: telemetry and audits remain comparable.
Tenant isolation: no cross-tenant leakage by design.
Kill switch is a first-class state: “off” is legitimate and auditable.

What we do not promise

No claim of “perfect safety” or “guaranteed compliance”.
No storage of raw end-user text by default.
No custom reason_code for a single customer (avoids consulting trap).
No dashboard-heavy surface area in v1.

Commercial overview

Software service

Insight Guard is offered as a hosted software service for teams evaluating runtime governance controls and contract-style AI decisioning.

Transparent policies

Pricing, terms, and privacy information are publicly available to support procurement review and platform verification workflows.

Support-ready

Customers can contact us for onboarding, access requests, commercial questions, and reference material through the contact details below.

Pricing Terms of Service Privacy Policy

FAQ

Do I need to point the domain to a VPS? No. This site is static and hosted via GitHub Pages; your API can stay elsewhere.
What makes this “contract-first”? Behavior is versioned, reason codes are stable, and every decision is auditable with replayable semantics.
Can this run “shadow-only”? Yes. You can log decisions without enforcing them, then turn on enforcement per tenant.
Is this “AI safety” marketing? No. This is runtime governance with auditable outputs and contract semantics.
Where can I read more? Notes live at /blog/.
Do you offer subscriptions? Yes. See /pricing/ for current commercial information.

Contact

If you want the reference pack, SDK examples, pricing details, or a contract-ready spec bundle, reach out below.

Product Insight Guard

Operator INNERSENSE LABS LLC

Support / Commercial Email info@insightguard.tech

Website https://insightguard.tech

Policies Pricing · Terms · Privacy

Reference Case (shadow)

SDK v0 (Node / Python)

Contract bundle

Ops runbooks

Read Notes →

Pricing · Terms · Privacy