TealTiger is the open-source runtime governance layer for AI agents. Enforce deterministic policy, control spend, and produce audit-ready evidence before an agent reaches a model, tool, or workflow.
Policy trace
Receipt
6
Framework adapters
11
Governance domains
12
LLM providers
0
External governance services
Part of the NVIDIA Inception ecosystem for AI startups.
Security, cost, identity, evidence, and workflow governance are treated as one runtime system instead of separate after-the-fact reviews.
FREEZE rules, PLAN_ONLY mode, hot-swap policy bundles, anti-tamper detection, and deterministic decisions without an LLM in the governance path.
Cryptographic receipts, reason codes, OpenTelemetry spans, SIEM exports, SARIF, JUnit, and JSON evidence.
Scope-bound access, Zero Standing Privilege, workload attestation, and revoked-agent enforcement.
PII detection, prompt-injection checks, content policy enforcement, Unicode normalization, and encoded output detection.
Budgets by request, session, agent, and day with anomaly alerts and reasoning-token ceilings.
OpenAI, Anthropic, Gemini, Bedrock, Azure OpenAI, Cohere, Mistral, and platform adapters.
Every policy decision is legible: a domain, a control ID, a risk reason, a release lineage, and evidence that can be reconstructed later.
| Domain | Control | Risk coverage | Since |
|---|---|---|---|
| Security | SEC | Secret leakage, PII exposure, prompt injection, content policy violations | v1.1 |
| Cost | COST | Budget overrun, cost anomaly, reasoning-token abuse, unattributed spend | v1.1 |
| Reliability | REL | Provider degradation, cascading failure, fallback chain failure | v1.1 |
| Memory | MEM | Data provenance gaps, instruction injection via memory, stored-context exfiltration | v1.2 |
| Registry | REG | Unapproved model or tool usage, MCP definition drift, supply chain compromise | v1.2 |
| Evidence | EVID | Audit gaps, non-repudiation failure, evidence tampering | v1.2 |
| Identity | NHI | Excessive standing privilege, missing attestation, scope creep | v1.3 |
| Workflow | FLOW | Policy bypass, missing approvals, weakened organization controls | v1.3 |
| Temporal | TEMP | Stale session context, off-hours execution, cooldown bypass | v1.3 |
| Drift | DRIFT | Behavioral deviation, model output regression, tool definition mutation | v1.3 |
| Observe | OBS | Passive monitoring gaps, visibility coverage, telemetry completeness | v1.4 |
Secret leakage, PII exposure, prompt injection, content policy violations
Since v1.1
Budget overrun, cost anomaly, reasoning-token abuse, unattributed spend
Since v1.1
Provider degradation, cascading failure, fallback chain failure
Since v1.1
Data provenance gaps, instruction injection via memory, stored-context exfiltration
Since v1.2
Unapproved model or tool usage, MCP definition drift, supply chain compromise
Since v1.2
Audit gaps, non-repudiation failure, evidence tampering
Since v1.2
Excessive standing privilege, missing attestation, scope creep
Since v1.3
Policy bypass, missing approvals, weakened organization controls
Since v1.3
Stale session context, off-hours execution, cooldown bypass
Since v1.3
Behavioral deviation, model output regression, tool definition mutation
Since v1.3
Passive monitoring gaps, visibility coverage, telemetry completeness
Since v1.4
TealTiger installs through standard developer channels. No hosted governance service is required for policy enforcement.
One-line adapter installs that add runtime governance to the frameworks your agents already use. No config changes to existing code.
The product story stays focused: governed AI systems for organizations that need runtime enforcement, cost accountability, and evidence.
Model risk management, regulatory audit trails, cost attribution, and separation of duties for agent-initiated transactions.
PHI governance in agent memory, consent-aware data handling, deterministic audit evidence, and role-scoped access.
Multi-agent orchestration governance, MCP supply-chain integrity, cost runaway prevention, and policy-as-code.
TealTiger highlights trusted contributors directly in the product story, giving sustained community work the visibility it deserves.
Recognized in TealTiger governance for sustained, high-quality work across SDK examples, guardrails, secret detection coverage, policy runtime features, tracing, SARIF export, and cross-SDK parity.
DeepSeek, CrewAI, MCP, Redis memory, and Python async examples
SaaS secret detector coverage and provider key detection
Custom guardrail support across TypeScript and Python SDKs
Policy hot-reload, OpenTelemetry tracing, SARIF export, and streaming evaluator work
Open-source contributor focused on practical AI agent tooling, repo readiness, workflow reliability, SDK examples, tests, and contributor documentation.
Provider quickstart examples across TealTiger SDK workflows
Policy examples and contributor-facing documentation
Focused SDK tests and metadata/link cleanup
Narrow, review-ready changes aligned with maintainer guidance
TealTiger is Apache 2.0. Contribute detection patterns, framework integrations, compliance mappings, and policy packs for real-world agent deployments.
Secret detection, prompt injection signatures, Unicode attack vectors, and incident-informed policy checks.
Governance adapters for LangChain, CrewAI, AG2, Haystack, Google ADK, Composio, Strands, PydanticAI, and more.
OWASP ASI mappings, industry templates, and control crosswalks for AI risk frameworks.
The first 25 contributors with merged PRs get permanent recognition and early access.
