TrigGuard
TRIGGUARD ARCHITECTURE

System architecture

TrigGuard is the single evaluation layer between automation and irreversible surfaces. It returns deterministic PERMIT / DENY / SILENCE and signs PERMIT outcomes—execution proceeds only when policy returns PERMIT.

TrigGuard operates at the point of irreversible execution, not just policy evaluation.

Protocol path (same content): /protocol/architecture · Execution authorization · Products · Docs · Pricing

Execution timeline

One action moving through the system—from request to verifiable receipt.

Control plane vs data plane

Governance and configuration above; real-time authorization on the hot path.

Fail-closed. Verifiable.

This is not monitoring. This is control before execution.

TrigGuard is not observability, coaching, or probabilistic scoring. It is a protocol-level gate: evaluate first, execute only on explicit PERMIT. TrigGuard operates at the point of irreversible execution, not just policy evaluation.

PERMIT / DENY / SILENCE

SILENCE means no authorization was issued—execution cannot proceed. DENY refuses explicitly; both enforce BLOCKED until a verified PERMIT exists.

Execution Authorization Boundary

Automation Systems
GitHub Actions · Terraform · GitLab CI · ArgoCD · Internal Systems
POST /execute · surface · action · context · idempotency_key
TRIGGUARD
Authoritative evaluation layer
policy-bound outcome

Managed service or private deployment—same fail-closed contract: no PERMIT, no execution.

signed receipt
Execution Surfaces
deploy.release · infra.apply · database.migrate · secrets.access · data.export

TrigGuard enforces the boundary between callers and irreversible surfaces. No evaluated request → no execution. No verified PERMIT → execution does not proceed.

Request → Gate → Verification → Receipt

Requests hit the gate first: policy evaluation, deterministic outcome, signed receipt on PERMIT—no bypass path.

Execution Lifecycle

Linear sequence of events from agent intent to signed proof. Each step is a gate; no bypass.

AGENT_INTENT
POST /execute
TRIGGUARD_DECISION
PERMIT / DENY / SILENCE
SYSTEM_ACTION
Execution Commit
SIGNED_RECEIPT
Cryptographic Proof

EXECUTION PROPERTIES

DETERMINISTIC_DECISION
Every execution request produces one deterministic authorization outcome: PERMIT, DENY, or SILENCE.
FAIL_CLOSED
If authorization cannot be confirmed, execution MUST NOT proceed.
UNREACHABLE_BEHAVIOR
If the TrigGuard authorization service is unreachable, clients MUST treat the path as blocked and abort execution—no fail-open retries.

Execution systems integrating TrigGuard must enforce authorization before commit. No action is performed without a valid authorization decision.

INTEGRATION SURFACES

How systems connect to the authorization boundary.

SURFACE_01

REST_API

For cloud-native agent frameworks.

SURFACE_02

NATIVE_SDK

Low-latency Go/Rust bindings for core systems.

SURFACE_03

SIDE_CAR

Proxy-based enforcement for legacy environments.

Architecture Diagram

The protocol creates a trust boundary between automation systems requesting execution and the surfaces where irreversible actions occur.

AUTOMATION SYSTEMS
GitHub Actions • Terraform • CI/CD • Internal Systems
POST /execute
{ action, context, metadata }
TRIGGUARD AUTHORIZATION LAYER
Policy Engine Decision Engine Receipt Generator
evaluate · deterministic · Ed25519 sign
signed receipt
{ decision, timestamp, signature }
CRYPTOGRAPHIC TRUST LAYER
Ed25519 signatures • Key discovery • Offline verification
GET /.well-known/trigguard-keys.json
verified execution
EXECUTION SURFACES
deploy.release infra.apply database.migrate secrets.access model.deploy data.export config.update billing.charge

Figure: TrigGuard deterministic execution flow and signed decision pipeline.

Execution Flow

Every authorization request follows a deterministic path through the system. The same inputs always produce the same decision.

01 Request Automation system sends authorization request to POST /execute with action type, context, and metadata.
02 Policy Evaluation Policy engine evaluates the request against configured rules and constraints—explicit conditions only.
03 Deterministic Decision Decision engine returns PERMIT, DENY, or SILENCE. Same inputs always produce the same output.
04 Receipt Signing Authorization decision is packaged into a receipt and signed with Ed25519 private key.
05 Response Signed receipt returned to client. Contains decision, timestamp, context hash, and cryptographic signature.
06 Verification Client may verify receipt offline using public key from /.well-known/trigguard-keys.json.
07 Execution If decision is PERMIT, automation system proceeds with execution. Receipt serves as audit proof.

[ SYSTEM_FAILURE_MODES · SEVERITY: HIGH ]

WHAT FAILS WITHOUT TRIGGUARD

Uncontrolled execution paths, no binding authorization at the gate.

  • AI agent executes payment without authorization

    No binding decision — money moves anyway.

  • CI pipeline deploys unintended change

    No commit control — production state is altered without validation.

  • Data export occurs without policy validation

    No execution boundary — sensitive data leaves the system.

Core Components

The protocol is composed of discrete components with clear responsibilities and interfaces.

authorization-api

Authorization API

REST endpoint accepting execution requests. Validates input schema, authenticates callers, routes to policy engine. Single entry point for all authorization decisions.

policy-engine

Policy Engine

Evaluates requests against configured rules: action/surface bindings, context constraints, time windows. Stateless, deterministic evaluation—no learned overrides.

decision-engine

Decision Engine

Deterministic decision logic. Given policy evaluation results, produces consistent authorization decision. No randomness, no external state dependency.

TrigGuard operates at the point of irreversible execution, not just policy evaluation.

receipt-generator

Receipt Generator

Packages authorization decision into structured receipt. Signs with Ed25519 private key. Includes timestamp, context hash, decision, and cryptographic signature.

key-discovery

Key Discovery Service

Serves public verification keys at /.well-known/trigguard-keys.json. Supports key rotation with validity periods. Enables offline verification.

verification-layer

Verification Layer

Client-side library for receipt verification. Fetches public keys, validates signatures, checks expiration. Works offline after initial key fetch.

Cryptographic Trust

TrigGuard uses cryptographic primitives to create verifiable, tamper-evident authorization records that work without network connectivity.

Signed Receipts

Every authorization decision produces a signed receipt. The receipt contains the decision, timestamp, action context hash, and an Ed25519 signature. Receipts cannot be forged or modified without detection. They serve as cryptographic proof that a specific authorization decision was made at a specific time.

Ed25519 Signatures

The protocol uses Ed25519 elliptic curve signatures. 256-bit security level. Fast signing and verification. Small key and signature sizes. Deterministic signature generation prevents timing attacks. Public keys are 32 bytes, signatures are 64 bytes.

Key Discovery

Public verification keys are served at a well-known URL: /.well-known/trigguard-keys.json. Keys include validity periods for rotation support. Clients cache keys and verify receipts locally. The discovery endpoint follows RFC 8615 conventions.

Offline Verification

Once a client has fetched the public key, receipts can be verified without network access. This enables air-gapped audit, local policy enforcement, and verification in disconnected environments. No callback to TrigGuard is required to validate a receipt.

Deterministic Execution Proof

The combination of deterministic decision logic and cryptographic signing creates verifiable proof of authorization. Given the same inputs, the system always produces the same decision. The signed receipt proves this decision was made by a TrigGuard-controlled private key.

Deployment Model

TrigGuard supports multiple deployment configurations depending on security requirements and infrastructure constraints.

AVAILABLE

TrigGuard Cloud

Hosted authorization service. API access via api.trigguardai.com. Managed key infrastructure, automatic rotation, high availability. Fastest path to integration.

AVAILABLE

CI/CD Integration

GitHub Actions, GitLab CI, Jenkins, CircleCI. Authorization checks integrated directly into deployment pipelines. Blocks or gates execution based on policy decisions.

AVAILABLE

API Gateway Layer

Deploy as middleware in API gateway. Intercepts requests to protected endpoints. Returns authorization decision before request reaches backend services.

PLANNED

Self-Hosted Runtime

On-premises deployment for enterprises requiring full control. Private key management, custom policy engine, air-gapped operation. Contact for enterprise discussion.

// SECURITY_GOVERNANCE

SOC2_ALIGNMENT

TG-01 protocol provides the immutable decision record required for Section CC7.2 (System Operations).

DATA_RESIDENCY

Metadata-only processing. No PII ingestion. Zero-retention authorization models available for local sidecars.

THREAT_MITIGATION

Non-repudiable execution. Protection against prompt-injection escalation and model-drift.

TrigGuard is designed to be the primary control point for AI-system assurance workflows. Every PERMIT / DENY / SILENCE decision is exportable to external SIEM/SOAR platforms.

DOWNLOAD_TECHNICAL_SPEC_PDF

// TECHNICAL_FAQ

Production behavior, failure mode, and verification. No marketing—SLA-level answers.

FAQ-01 How does TrigGuard hit its latency targets?
RESPONSE: Two scoped numbers: kernel hot path <5ms p99 and end-to-end evaluation <15ms p99 on the finite-state path under normal load. A lightweight, compiled policy engine and localized caching of cryptographic keys avoid model-inference round-trips on the critical path.
FAQ-02 What happens if the TrigGuard service is unreachable?
RESPONSE: The system is Fail-Closed by default. If the authorization boundary cannot verify a signed permit, the execution is killed. For high-availability environments, we provide local sidecar deployments to ensure line-rate authorization even during network partitions.
FAQ-03 Does TrigGuard read my application data?
RESPONSE: No. TrigGuard operates on Action Metadata and Intent Schemas. We do not ingest your raw database contents or full chat histories. We only evaluate the specific parameters of the irreversible action being requested (e.g., destination, amount, system-level impact).
FAQ-04 How is the ‘Signed Receipt’ verified offline?
RESPONSE: Every decision is signed using Ed25519 asymmetric cryptography. Your execution surface can verify the authenticity of a permit using our public keys (available via /.well-known/trigguard-keys.json) without ever making an outbound call to our servers.