Agent ID
A unique identifier for an AI agent within a TealTiger-governed system. Used in execution identity to scope policies, audit trails, and memory access to a specific agent instance.BundleExporter
A v1.2 component that exports governance evidence in structured formats. Supports SARIF v2.1.0, JUnit XML, and JSON output. Used to feed evidence into CI/CD pipelines, SIEM systems, and compliance dashboards. Related: TEEC, TealAuditCircuit Breaker
A reliability pattern implemented by TealCircuit and TealReliability. Tracks consecutive failures to a downstream provider and transitions through three states:- CLOSED — Normal operation. Requests pass through.
- OPEN — Too many failures. Requests are blocked (reason code:
CIRCUIT_OPEN). - HALF_OPEN — Testing recovery. A limited number of requests are allowed through.
Classification (Memory)
A data sensitivity label applied to memory entries in TealMemory. Four levels:
An agent can only read memory at or below its own classification level.
Related: TealMemory, Scope
Correlation ID
A UUID v4 string that uniquely identifies a single governance evaluation request. Auto-generated by TealEngine and propagated through all modules, audit events, and evidence envelopes. Compatible with OpenTelemetry trace IDs. Related: Execution Identity, TealAuditDecision
The deterministic output of a TealEngine evaluation. A typed object containing:action— What to do (DecisionAction)reason_codes— Why (Reason Code)risk_score— How severe (0–100)mode— Enforcement mode (Policy Mode)correlation_id— Request trace IDpolicy_id/policy_version— Which policy was evaluatedcomponent_versions— SDK and engine versions
Decision Action
The enforcement verb in a Decision. Tells your code what to do with the request. See DecisionAction enum values for the full list. Related: Error Reference — Decision ActionsDecisionAction (Enum Values)
The 12 possible actions returned inDecision.action:
When multiple modules return different actions, most restrictive wins.
Deterministic Governance
TealTiger’s core design principle: governance decisions are made by deterministic code (policy rules, pattern matching, threshold checks) — never by an LLM. This guarantees reproducible, auditable, and testable outcomes. Given the same input and policy, the same decision is always returned. Related: Decision Philosophy, Security vs GovernanceEvidence Envelope
A structured record produced by TealAudit for each governance evaluation. Contains the Decision, all reason codes, module outputs, timestamps, and correlation IDs. Evidence envelopes are the atomic unit of the audit trail and can be exported via BundleExporter. Related: Audit Event Schema, TEECFail-Closed
The default failure behavior in TealEngineV12. If a governance module throws an exception during evaluation, the engine returnsDENY rather than allowing an ungoverned request through. Configurable via failurePolicy.default in engine options.
Opposite: Fail-Open
Fail-Open
An alternative failure behavior where module failures result inALLOW. Use with caution — this means ungoverned requests can pass through during outages. Configured by setting failurePolicy.default: 'FAIL_OPEN'.
Opposite: Fail-Closed
Governance Domain
A category of governance risk. TealTiger v1.3 covers 10 domains:Guardrail
A runtime check that evaluates content, tool usage, or agent behavior against a policy. Guardrails are implemented as TealModules and produce ModuleResults that feed into the engine’s merge logic. Related: Guardrail InternalsModuleContext
The context object passed to every TealModule during evaluation. Contains:correlation_id— Request trace IDtenant_id— Multi-tenant identifier (optional)user_id— End user identifier (optional)session_id— Session identifier (optional)agent_id— Agent identifier (optional)policy_version— Policy version being evaluatedteec_version— TEEC registry versiontimestamp— Evaluation start time (Unix ms)
ModuleResult
The output of a TealModule evaluation. Contains:action— The module’s recommended DecisionActionreason_codes— Array of reason codes explaining the resultevent_type— TEEC event type for auditfindings— Secret findings (TealSecrets only)metadata— Additional module-specific data
Most Restrictive Wins
The merge strategy used by TealEngineV12 when combining results from multiple modules. Each DecisionAction has a severity score (0–100). The action with the highest severity becomes the final decision. For example, if TealGuard returnsALLOW (0) and TealSecrets returns DENY (100), the final action is DENY.
Related: Policy Evaluation Logic
Policy Mode (ENFORCE, MONITOR, REPORT_ONLY)
Controls how the engine acts on governance decisions:
Modes can be set globally, per environment, or per policy. Priority: policy-specific > environment-specific > global.
Related: Policy Modes, Phased Adoption
Reason Code
A machine-readable string explaining why a Decision was made. TealTiger v1.2 defines 32 reason codes across 8 categories. Every decision contains at least one reason code. Related: Error Reference — Reason Codes, Reason CodesRisk Score
A numeric value between 0 and 100 (inclusive) representing the severity of a governance decision. Computed from the DecisionAction severity:ALLOW = 0, TRANSFORM = 50, DEGRADE = 60, REDACT = 70, REQUIRE_APPROVAL = 80, DENY = 100.
Related: Risk Scores
SARIF
Static Analysis Results Interchange Format (v2.1.0). An OASIS standard JSON format for expressing static analysis results. BundleExporter can export governance evidence as SARIF for integration with GitHub Code Scanning, Azure DevOps, and other SARIF-compatible tools.Scope (Memory)
The boundary within which memory is accessible in TealMemory. Five scopes:
An agent can only access memory within its configured scope. Cross-scope access triggers
MEMORY_SCOPE_VIOLATION.
Sidecar
A deployment pattern where TealTiger runs as a separate process alongside your application (e.g., as a Kubernetes sidecar container). The sidecar exposes an HTTP API that your application calls for governance decisions, decoupling governance from application code. Related: Deployment Overview, KubernetesTealAudit
The audit logging component. Produces versioned evidence envelopes with automatic PII redaction. Supports structured logging, event correlation, and export to external systems. Related: Audit & Redaction, API ReferenceTealCircuit
The v1.1 circuit breaker component. Prevents cascading failures by tracking error rates to downstream providers and opening the circuit when a threshold is exceeded. Superseded in v1.2 by TealReliability, which adds retry budgets and fallback chains. Related: API ReferenceTealEngine / TealEngineV12
The core orchestration layer. Evaluates requests against policies by dispatching to registered TealModules.- TealEngine (v1.1) — Sequential evaluation, single-module focus.
- TealEngineV12 (v1.2) — Parallel evaluation (
Promise.allSettled), multi-module merge using most restrictive wins, fail-closed defaults, and TEEC validation.
TealGuard
The content safety module. Provides client-side guardrails for PII detection, prompt injection detection, content moderation, and unsafe code detection. Returns reason codes likePII_DETECTED, PROMPT_INJECTION_DETECTED, HARMFUL_CONTENT_DETECTED, and UNSAFE_CODE_DETECTED.
Related: Guardrail Internals, API Reference
TealMemory
The v1.2 memory governance module. Controls what can be written to and read from agent memory using scope boundaries and classification levels. Supports five scopes and four classification levels. Related: API ReferenceTealModule
The interface that all governance modules implement. Requires:name— Module identifierversion— Semantic versionevaluate(request, ctx, policy)— Core evaluation method returning a ModuleResultinit(config)— Optional lazy initializationdestroy()— Optional cleanup
TealRegistry
The v1.2 model and tool registry module. Maintains allowlists of approved models and tools with provenance verification. ReturnsTOOL_NOT_ALLOWED or MODEL_NOT_ALLOWLISTED when an unregistered model or tool is used.
Related: API Reference
TealReliability
The v1.2 reliability module. Provides retry budgets, circuit breakers, fallback chains, and degradation strategies. Supersedes TealCircuit with a more comprehensive reliability model. Related: API ReferenceTealSecrets
The v1.2 secret detection module. Scans content for 500+ secret patterns (API keys, tokens, private keys, credentials) with confidence scoring. Returns findings withfinding_id, type, confidence, and fingerprint.
Related: API Reference
TEEC (Typed Evidence & Evidence Contract)
The formal evidence contract. TEEC v2.0.0 (shipped in v1.3) defines:- 60+ reason codes — Why a decision was made
- 30+ event types — What happened
- 16 decision actions — What to do
- NHI identity fields — Who triggered the action
- Cryptographic proof — Merkle root, inclusion proof, RFC 3161 anchor
- Automation level — What governance automation was applied
- Control ID — Which governance control was triggered
- OWASP category — Which ASI risk is addressed
v1.3 Terms
Automation Level
The governance automation applied to a policy rule. Four levels:
Related: Automation Levels
FREEZE Rule
An immutable, non-overridable safety control. FREEZE rules:- Are evaluated FIRST in the pre-evaluation pipeline
- Cannot be modified, disabled, or removed by application code
- Persist across policy hot-swaps and process restarts
- Log
FREEZE_TAMPER_ATTEMPTon any modification attempt
Governance Passport
A rolling cryptographic attestation proving continuous governance coverage over a time period. Generated by TealProof. Contains sealed Merkle trees for each time window, enabling third-party verification that an agent was continuously governed. Related: TealProofJIT Grant (Just-In-Time)
A time-bounded, scope-limited access grant for an agent under Zero Standing Privilege. JIT grants expire automatically and are audited. No agent holds permanent elevated permissions. Related: NHI GovernanceNon-Human Identity (NHI)
An AI agent treated as a first-class principal with identity, lifecycle, scope, and attestation. NHIs have three lifecycle states:active, suspended, revoked. Governed by the Identity (NHI) domain.
Related: NHI Governance
PLAN_ONLY Mode
A governance mode that evaluates what the decision would be without executing enforcement or emitting evidence. Used for agent planning loops and UI previews. Side-effecting actions returnDENY with PLAN_ONLY_BLOCK.
Policy Bundle
A signed, versioned package of governance policies distributed by the governance team. Bundles include policies, FREEZE rules, cost limits, and NHI configuration. Verified via Ed25519 signature before loading. Related: Governance at Scale, ConfigurationSeparation of Duties
The enterprise operating model where governance teams define policy (author, sign, publish bundles) and development teams consume the SDK (integrate, call evaluate, handle decisions). Developers cannot modify or bypass governance controls. Related: Governance at ScaleTealClassifier
v1.3 local ML inference module. Performs content classification using ONNX models without external API calls. Supports 4 ensemble modes combining regex with ML for high-accuracy detection. Related: API ReferenceTealDrift
v1.3 behavioral drift detection module. Monitors agent behavior against statistical baselines and alerts when behavior deviates beyond configured thresholds. Related: API ReferenceTealEngineV13
The v1.3 core engine. Extends TealEngineV12 with pre-evaluation stages (FREEZE, NHI, temporal), post-evaluation hooks (evidence, SIEM, response hooks), automation levels, and PLAN_ONLY mode. Backward-compatible —evaluateV12() preserved.
Related: API Reference