HELM Framework Integration Guide

How frameworks embed HELM governance using the Portable Effect Model.

Framework Index

Framework	Language	Integration Pattern	Adapter Status
OpenAI Agents SDK	Python	Middleware	Reference
OpenAI Agents SDK	JS/TS	Middleware	Reference
Google ADK	Python	Tool wrapper	Reference
LangGraph	Python	Node wrapper	Reference
LangChain	Python	Callback handler	Reference
Mastra	TypeScript	Middleware	Reference
CrewAI	Python	Tool wrapper	Reference
AI SDK (generic)	TypeScript	Middleware	Reference
Microsoft Agent Framework	.NET	Middleware	Planned
AutoGen	Python	Tool wrapper	Planned
DSPy	Python	Module wrapper	Planned
Semantic Kernel	C#	Filter	Planned

---

Integration Patterns

Pattern 1: OpenAI Agents SDK (Python)

from openai import agents
from helm_sdk import HelmGovernance

# Create governed agent
governance = HelmGovernance(
    endpoint="http://localhost:4001",
    preset="engineering",
    jurisdiction="eu-gdpr",
)

agent = agents.Agent(
    name="code-assistant",
    model="gpt-4o",
    tools=[...],
    middleware=[governance.middleware()],
)

# Every tool call automatically goes through HELM
result = agent.run("Deploy the new version")
# → HELM intercepts → PDP evaluates → ALLOW/DENY/ESCALATE

Pattern 2: OpenAI Agents SDK (JS/TS)

import { Agent } from 'openai/agents';
import { HelmGovernance } from '@mindburn/helm';

const governance = new HelmGovernance({
  endpoint: 'http://localhost:4001',
  preset: 'engineering',
});

const agent = new Agent({
  name: 'code-assistant',
  model: 'gpt-4o',
  tools: [...],
  middleware: [governance.middleware()],
});

Pattern 3: Google ADK

from google.adk import Agent
from helm_sdk import HelmGovernance

governance = HelmGovernance(
    endpoint="http://localhost:4001",
    auth_mode="google_oauth",
)

agent = Agent(
    model="gemini-2.0-flash",
    tools=[governance.wrap(tool) for tool in my_tools],
)

Pattern 4: LangGraph

from langgraph.graph import StateGraph
from helm_sdk import HelmGovernance

governance = HelmGovernance(endpoint="http://localhost:4001")

def governed_tool_node(state):
    tool_call = state["tool_calls"][-1]
    result = governance.submit_effect(
        tool_call["name"],
        tool_call["args"],
    )
    if result.verdict == "DENY":
        return {"denied": True, "reason": result.reason}
    # Execute tool and complete
    output = execute_tool(tool_call)
    governance.complete(result.intent_id, success=True)
    return {"output": output}

graph = StateGraph(...)
graph.add_node("tools", governed_tool_node)

Pattern 5: Mastra (TypeScript)

import { Agent } from "mastra";
import { HelmGovernance } from "@mindburn/helm";

const governance = new HelmGovernance({
  endpoint: "http://localhost:4001",
  preset: "engineering",
});

const agent = new Agent({
  name: "mastra-agent",
  model: openai("gpt-4o"),
  tools: governance.wrapAll(myTools),
});

Pattern 6: CrewAI

from crewai import Agent, Task, Crew
from helm_sdk import HelmGovernance

governance = HelmGovernance(endpoint="http://localhost:4001")

agent = Agent(
    role="analyst",
    goal="Analyze data",
    tools=[governance.wrap(tool) for tool in tools],
)

Pattern 7: AI SDK (generic)

import { generateText } from "ai";
import { HelmGovernance } from "@mindburn/helm";

const governance = new HelmGovernance({ endpoint: "http://localhost:4001" });

const result = await generateText({
  model: openai("gpt-4o"),
  tools: governance.wrapAll(myTools),
  prompt: "Analyze the sales data",
});

---

Framework Adapter Structure

Each framework adapter follows this structure:

sdk/{language}/src/adapters/{framework}/
├── adapter.{ext}       # Core adapter implementation
├── middleware.{ext}     # Middleware/interceptor (if pattern A/B)
├── wrapper.{ext}       # Tool wrapper (if pattern C)
└── README.md           # Framework-specific guide

Testing Framework Integrations

# Run framework integration tests
helm test frameworks --framework openai-agents
helm test frameworks --framework langgraph
helm test frameworks --all