🛡️ @aegis-kernel/openai

OpenAI Function Calling & Assistant Tools Deterministic Safety Guard
Pre-Execution Clearance • Structured Self-Healing Error Feedback • Tool Message Interception

🚀 Overview

@aegis-kernel/openai intercepts OpenAI chat completion tool calls (message.tool_calls) and enforces deterministic policy invariants before dispatching tool execution. If a tool call violates an invariant, the guard automatically synthesizes a self-healing tool error message that guides the LLM to auto-correct its arguments.

📦 Installation

npm install @aegis-kernel/openai @aegis-kernel/core

⚡ Quickstart

import OpenAI from 'openai';
import { AegisOpenAIGuard } from '@aegis-kernel/openai';

const openai = new OpenAI();
const guard = new AegisOpenAIGuard({
  mode: 'enforce',
  packs: ['@aegis/sql-guard', '@aegis/finance-guard'],
});

const completion = await openai.chat.completions.create({
  model: 'gpt-4o',
  messages: [{ role: 'user', content: 'Clean up inactive users table' }],
  tools: myToolDefinitions,
});

const message = completion.choices[0].message;

if (message.tool_calls) {
  for (const toolCall of message.tool_calls) {
    // Automatically intercepts, validates invariants, and executes safely
    const toolMessage = await guard.handleToolCall(toolCall, async (args) => {
      return await myLocalToolExecutor(args);
    });

    // toolMessage contains either the execution output or structured self-healing feedback
    messages.push(toolMessage);
  }
}

📄 License

Distributed under the MIT License. Copyright (c) 2026 Sneh Gabani.