Langfuse Integration

Integrate TealTiger with Langfuse to combine governance decisions with LLM observability and tracing.

Why integrate TealTiger with Langfuse?

Langfuse provides LLM-specific observability. Adding TealTiger gives you:

Governance visibility - See policy decisions alongside LLM traces
Cost attribution - Link TealTiger costs to Langfuse traces
Debug workflows - Understand why requests were blocked
Compliance evidence - Audit trails in Langfuse

Quick start

Install both packages:

npm install tealtiger langfuse
# or
pip install tealtiger langfuse

Configure TealTiger to export to Langfuse:

import { TealTiger } from 'tealtiger';
import { Langfuse } from 'langfuse';

// Initialize Langfuse
const langfuse = new Langfuse({
  publicKey: process.env.LANGFUSE_PUBLIC_KEY,
  secretKey: process.env.LANGFUSE_SECRET_KEY
});

// Configure TealTiger to export to Langfuse
const teal = new TealTiger({
  policies: { /* ... */ },
  telemetry: {
    langfuse: {
      enabled: true,
      client: langfuse,
      exportDecisions: true,
      exportCosts: true
    }
  }
});

// Use TealTiger - decisions automatically exported
const decision = await teal.evaluate(request);

from tealtiger import TealTiger
from langfuse import Langfuse

# Initialize Langfuse
langfuse = Langfuse(
    public_key=os.environ["LANGFUSE_PUBLIC_KEY"],
    secret_key=os.environ["LANGFUSE_SECRET_KEY"]
)

# Configure TealTiger to export to Langfuse
teal = TealTiger({
    "policies": {},
    "telemetry": {
        "langfuse": {
            "enabled": True,
            "client": langfuse,
            "exportDecisions": True,
            "exportCosts": True
        }
    }
})

# Use TealTiger - decisions automatically exported
decision = await teal.evaluate(request)

What gets exported?

TealTiger exports governance data as Langfuse observations:

Decision observations

{
  "type": "event",
  "name": "tealtiger.decision",
  "metadata": {
    "action": "DENY",
    "reason_codes": ["TOOL_NOT_ALLOWED"],
    "risk_score": 85,
    "policy_id": "security.tool_access.v1",
    "mode": "ENFORCE"
  },
  "level": "WARNING"
}

Cost observations

{
  "type": "event",
  "name": "tealtiger.cost",
  "metadata": {
    "estimated_cost": 0.05,
    "tokens_input": 1000,
    "tokens_output": 500,
    "model": "gpt-4"
  }
}

Complete example

import { TealTiger } from 'tealtiger';
import { Langfuse } from 'langfuse';
import { ChatOpenAI } from 'langchain/chat_models/openai';

// Initialize Langfuse
const langfuse = new Langfuse({
  publicKey: process.env.LANGFUSE_PUBLIC_KEY,
  secretKey: process.env.LANGFUSE_SECRET_KEY
});

// Initialize TealTiger with Langfuse export
const teal = new TealTiger({
  policies: {
    tools: {
      web_search: { allowed: true },
      file_delete: { allowed: false }
    },
    budget: {
      maxCostPerRequest: 0.50
    }
  },
  telemetry: {
    langfuse: {
      enabled: true,
      client: langfuse,
      exportDecisions: true,
      exportCosts: true
    }
  }
});

// Create a trace in Langfuse
const trace = langfuse.trace({
  name: "agent-execution",
  userId: "user-123"
});

// Evaluate with TealTiger (exports to Langfuse)
const decision = await teal.evaluate({
  action: 'tool.execute',
  tool: 'web_search',
  context: {
    traceId: trace.id  // Link to Langfuse trace
  }
});

// Add LLM call to trace
const generation = trace.generation({
  name: "llm-call",
  model: "gpt-4",
  input: "Search for AI news"
});

// Execute if allowed
if (decision.action === 'ALLOW') {
  const model = new ChatOpenAI();
  const response = await model.call([
    { role: 'user', content: 'Search for AI news' }
  ]);
  
  generation.end({
    output: response.content
  });
}

// Finalize trace
await langfuse.flushAsync();

from tealtiger import TealTiger
from langfuse import Langfuse
from langchain.chat_models import ChatOpenAI

# Initialize Langfuse
langfuse = Langfuse(
    public_key=os.environ["LANGFUSE_PUBLIC_KEY"],
    secret_key=os.environ["LANGFUSE_SECRET_KEY"]
)

# Initialize TealTiger with Langfuse export
teal = TealTiger({
    "policies": {
        "tools": {
            "web_search": {"allowed": True},
            "file_delete": {"allowed": False}
        },
        "budget": {
            "maxCostPerRequest": 0.50
        }
    },
    "telemetry": {
        "langfuse": {
            "enabled": True,
            "client": langfuse,
            "exportDecisions": True,
            "exportCosts": True
        }
    }
})

# Create a trace in Langfuse
trace = langfuse.trace(
    name="agent-execution",
    user_id="user-123"
)

# Evaluate with TealTiger (exports to Langfuse)
decision = await teal.evaluate({
    "action": "tool.execute",
    "tool": "web_search",
    "context": {
        "traceId": trace.id
    }
})

# Add LLM call to trace
generation = trace.generation(
    name="llm-call",
    model="gpt-4",
    input="Search for AI news"
)

# Execute if allowed
if decision["action"] == "ALLOW":
    model = ChatOpenAI()
    response = model.invoke([
        {"role": "user", "content": "Search for AI news"}
    ])
    
    generation.end(output=response.content)

# Finalize trace
langfuse.flush()

Viewing in Langfuse

TealTiger decisions appear in Langfuse traces:

Trace view - See governance decisions in the trace timeline
Metadata - Policy details, reason codes, risk scores
Cost tracking - TealTiger costs alongside LLM costs
Filtering - Filter traces by decision action or policy

Best practices

Link traces - Use traceId to connect TealTiger and Langfuse
Export costs - Track governance overhead
Use metadata - Add context to decisions
Flush regularly - Ensure data is sent to Langfuse
Monitor denials - Create Langfuse dashboards for blocked requests

Next steps

LangSmith integration

Alternative LLM observability

OpenTelemetry

Standard observability protocol

Audit schema

Learn about audit events

Other integrations

View all integrations

Getting Started

Core Concepts

Governance Domains

Policy Authoring

Architecture

Integrations

Deployment

API Reference

Cookbook

Compliance & GRC

Guides

Reference

Versions

Playground

Dashboard

Blog

Why integrate TealTiger with Langfuse?

Quick start

What gets exported?

Decision observations

Cost observations

Complete example

Viewing in Langfuse

Best practices

Next steps

LangSmith integration

OpenTelemetry

Audit schema

Other integrations

​Why integrate TealTiger with Langfuse?

​Quick start

​What gets exported?

​Decision observations

​Cost observations

​Complete example

​Viewing in Langfuse

​Best practices

​Next steps

LangSmith integration

OpenTelemetry

Audit schema

Other integrations

Why integrate TealTiger with Langfuse?

Quick start

What gets exported?

Decision observations

Cost observations

Complete example

Viewing in Langfuse

Best practices

Next steps