Adding Governance Guardrails to Haystack 3.0 Pipelines with TealTiger

TealTiger, an open-source governance engine, has been integrated into Haystack 3.0 pipelines as a custom component, providing deterministic policy enforcement with sub-2ms evaluation and structured audit evidence. The integration allows developers to add guardrails that block PII, enforce cost limits, and restrict tool usage before requests reach LLMs, addressing compliance needs in regulated industries.

Haystack 3.0 redesigned everything around composable pipelines — you connect components like LEGO bricks to build RAG, chat, and agent workflows. But once those pipelines hit production, you need answers to questions like: TealTiger https://github.com/agentguard-ai/tealtiger is an open-source governance engine that answers these deterministically — no LLM in the governance path, sub-2ms evaluation, structured audit evidence. This post shows how to wire TealTiger into a Haystack 3.0 pipeline as a custom component. Haystack pipelines are powerful but trust-everything by default. A ChatGenerator will happily pass PII to OpenAI. A ToolInvoker will execute any tool the LLM requests. In regulated environments healthcare, finance, government , that's a compliance violation waiting to happen. TealTiger adds a governance layer that: pip install tealtiger-haystack No separate adapter package needed — TealTiger works directly as a Haystack custom component. Haystack 3.0's @component decorator makes this straightforward. We create a TealTigerGuard component that sits in the pipeline between user input and the LLM: python from haystack import component, Pipeline from haystack.components.generators.chat import OpenAIChatGenerator from haystack.dataclasses import ChatMessage from tealtiger import TealTiger from tealtiger.core.engine.types import PolicyMode @component class TealTigerGuard: """Governance guardrail component for Haystack 3.0 pipelines.""" def init self, policies: dict, mode: str = "ENFORCE", agent id: str = "haystack-agent", : self.engine = TealTiger policies=policies, mode=PolicyMode mode , agent id=agent id, @component.output types messages=list, List ChatMessage — passed through if allowed blocked=bool, decision=dict, def run self, messages: list : Extract text from the last user message user text = "" for msg in reversed messages : if msg.role.value == "user": user text = msg.content break Evaluate governance decision = self.engine.evaluate content=user text, tool name=None, metadata={"pipeline": "haystack", "component": "TealTigerGuard"}, if decision.action == "DENY": return { "messages": , "blocked": True, "decision": { "action": decision.action, "reason codes": str rc for rc in decision.reason codes , "risk score": decision.risk score, }, } return { "messages": messages, "blocked": False, "decision": { "action": "ALLOW", "reason codes": "POLICY COMPLIANT" , "risk score": 0, }, } Here's a complete pipeline that scans user queries before they reach the LLM: python from haystack import Pipeline from haystack.components.generators.chat import OpenAIChatGenerator from haystack.dataclasses import ChatMessage Define governance policies policies = { "pii block": { "enabled": True, "categories": "ssn", "credit card", "email", "phone" , }, "cost limit": { "enabled": True, "max per session": 0.50, $0.50 per session }, "tool allowlist": { "enabled": True, "allowed": "search", "lookup ", "calculate" , }, } Create pipeline pipe = Pipeline pipe.add component "governance", TealTigerGuard policies=policies, mode="ENFORCE" pipe.add component "llm", OpenAIChatGenerator model="gpt-4o-mini" Connect: governance output → LLM input only if not blocked pipe.connect "governance.messages", "llm.messages" Run with a safe query result = pipe.run { "governance": { "messages": ChatMessage.from user "What is the capital of France?" } } print result "llm" "replies" 0 .content → "The capital of France is Paris." Run with PII — gets blocked before reaching OpenAI result = pipe.run { "governance": { "messages": ChatMessage.from user "My SSN is 123-45-6789, look up my records" } } print result "governance" "blocked" True print result "governance" "decision" "reason codes" "PII DETECTED" LLM never sees the SSN For agentic pipelines where the LLM calls tools, you can wrap the tool execution step: @component class TealTigerToolGuard: """Guards tool invocations in agent pipelines.""" def init self, policies: dict, mode: str = "ENFORCE" : self.engine = TealTiger policies=policies, mode=PolicyMode mode @component.output types allowed=bool, decision=dict def run self, tool name: str, tool args: dict : decision = self.engine.evaluate content=str tool args , tool name=tool name, return { "allowed": decision.action == "ALLOW", "decision": { "action": decision.action, "risk score": decision.risk score, "reason codes": str rc for rc in decision.reason codes , "tool name": tool name, }, } TealTiger supports three modes that map to deployment stages: | Mode | Behavior | Use Case | |---|---|---| ENFORCE | Blocks violations | Production with strict compliance | MONITOR | Logs violations but allows through | Staging / shadow mode | REPORT ONLY | Skips evaluation, always allows | Development / dry-run | Shadow mode — see what would be blocked without breaking anything guard = TealTigerGuard policies=policies, mode="MONITOR" Every governance decision produces a structured receipt: { "decision id": "550e8400-e29b-41d4-a716-446655440000", "action": "DENY", "risk score": 85, "reason codes": "PII DETECTED:ssn" , "policy id": "pii block", "evaluation time ms": 0.8, "agent id": "haystack-agent", "correlation id": "trace-abc-123", "timestamp": "2026-08-17T10:30:00Z" } This feeds directly into SOC2/HIPAA compliance workflows — no manual log parsing. TealTiger's governance path is deterministic regex + fnmatch, no LLM calls : For comparison, a single LLM call takes 500-3000ms. Governance adds negligible latency. | TealTiger | Custom validators | No governance | | |---|---|---|---| | PII detection | 40+ patterns, zero config | Write your own regex | ❌ | | Tool allowlisting | Built-in with glob patterns | Write your own | ❌ | | Cost tracking | Per-request with budgets | DIY with token counters | ❌ | | Audit receipts | Structured TEEC format | DIY JSON | ❌ | | Governance modes | ENFORCE/MONITOR/REPORT ONLY | DIY | ❌ | | Framework lock-in | None works anywhere | Haystack-specific | N/A | pip install tealtiger haystack-ai TealTiger also integrates with LangChain https://pypi.org/project/langchain-tealtiger/ , AG2 https://docs.ag2.ai/extensions/tealtiger , and MLflow https://github.com/agentguard-ai/tealtiger/tree/main/packages/mlflow-tealtiger — same governance engine, different frameworks. TealTiger is Apache 2.0 licensed. We're an NVIDIA Inception member building deterministic governance for AI agents.