langchain-omega

LangChain Integration for OmegaEngine

Documentation • Playground • Issues

✨ Features

• 🔗 Callback Handler — Log all LLM decisions to OmegaEngine
• 🤖 Agent Wrapper — Enforce policies on LangChain agents
• 📊 Decision Logging — Automatic audit trail for all interactions
• 🛡️ Safety Guards — Pre/post-processing safety filters
• ⚡ Async Support — Full async/await compatibility

📦 Installation

pip install langchain-omega

🚀 Quick Start

Callback Handler

from langchain_openai import ChatOpenAI
from langchain_omega import OmegaCallbackHandler

# Create callback handler
omega_handler = OmegaCallbackHandler(
    api_key="your-omega-api-key",
    policy_id="pol_abc123",
)

# Use with any LangChain LLM
llm = ChatOpenAI(callbacks=[omega_handler])

# All invocations are now logged to OmegaEngine
response = llm.invoke("What is the capital of France?")

Decision Tool

from langchain.agents import create_openai_functions_agent
from langchain_omega import OmegaDecisionTool

# Add decision tool to your agent
decision_tool = OmegaDecisionTool(
    api_key="your-omega-api-key",
    domain="finance",
)

tools = [decision_tool, ...other_tools]
agent = create_openai_functions_agent(llm, tools, prompt)

# Agent can now make governed decisions
result = agent.invoke({
    "input": "Should we approve this loan application?"
})

Safety Guard Chain

from langchain_omega import GovernedLLMChain

# Wrap any chain with safety guards
safe_chain = GovernedLLMChain(
    chain=your_chain,
    api_key="your-omega-api-key",
    policy_id="safety_policy",
    block_on_deny=True,
)

# Chain will block unsafe outputs
result = safe_chain.invoke({"input": "..."})

📊 What Gets Logged

Every LLM interaction logs:

• ✅ Input prompt
• ✅ Output response
• ✅ Token usage
• ✅ Latency
• ✅ Model name
• ✅ Policy evaluation result
• ✅ Risk assessment
• ✅ Audit trail ID

📄 License

Licensed under the Apache License 2.0.

Built with ❤️ by the OmegaEngine team