OpenAI Integration

Guard OpenAI Function Calls Before They Run

Add approval gates and blocking to GPT-4 function calls before they execute. No more silent tool execution.

GPT-4 function calls execute silently

No native way to pause, block, or require human review. When GPT-4 decides to call a function, it happens automatically. You only find out after the fact — after the email is sent, the record is deleted, or the API is called.

Without Runplane

Agent decidesTool executesNo control

Intercept function calls before execution

Receive the function_call from OpenAI, pass it through Runplane for policy evaluation, then execute only if approved. Full control over every GPT function call.

agent.ts

import { Shield } from "@runplane/runplane-sdk";

const runplane = new Shield({
  apiKey: process.env.RUNPLANE_API_KEY
});

// After receiving function_call from OpenAI, before executing
const message = response.choices[0].message;

if (message.function_call) {
  const result = await runplane.guard(
    message.function_call.name,
    "openai-agent",
    JSON.parse(message.function_call.arguments),
    async () => executeFunctionCall(message.function_call)
  );

  if (result.decision === "BLOCKED") {
    // Handle blocked action
    return { error: "Function call blocked by policy" };
  }
}

What You Get

Block dangerous function calls

Require human approval for sensitive operations

Full audit trail of GPT actions

How Runplane Works

Intercept

guard() intercepts the action before execution

Decide

Policy engine evaluates and returns a decision

Execute or Halt

Action runs, blocks, or waits for approval

ALLOW

REQUIRE_APPROVAL

BLOCK

Control Any Function Call

send_email()

delete_record()

create_user()

transfer_funds()

Start free at runplane.ai

Add runtime control to your OpenAI agents in minutes. No credit card required.