DEV Community

Anton Illarionov
Anton Illarionov

Posted on

OpenAI Agents SDK + ODEI: Constitutional World Model Integration

#ai #openai #agents

OpenAI Agents SDK + ODEI

OpenAI's Agents SDK handles tool calling and handoffs. ODEI adds constitutional governance and persistent memory.

Integration 

from agents import Agent, Runner, function_tool
import requests

ODEI_TOKEN = "your-token"

@function_tool
def check_action(action: str) -> dict:
    return requests.post(
        "https://api.odei.ai/api/v2/guardrail/check",
        headers={"Authorization": f"Bearer {ODEI_TOKEN}"},
        json={"action": action, "severity": "medium"}
    ).json()

@function_tool  
def query_memory(term: str) -> dict:
    return requests.post(
        "https://api.odei.ai/api/v2/world-model/query",
        headers={"Authorization": f"Bearer {ODEI_TOKEN}"},
        json={"queryType": "search", "searchTerm": term}
    ).json()

agent = Agent(
    name="constitutional_agent",
    instructions="Always check_action before consequential operations.",
    tools=[check_action, query_memory]
)
Enter fullscreen mode Exit fullscreen mode

The Pattern

  1. Session start: query_memory("context") - inject world model state
  2. Before tools: check_action(call) - constitutional validation
  3. World model persists across sessions automatically

Production

With this pattern: 0 duplicate executions, 0 unauthorized ops, clean session resumption.

Top comments (0)