@agentiam/core

The core policy engine, evaluation logic, and authorization gateway for Agent IAM.

@agentiam/core provides the foundational structures to build human-in-the-loop and policy-driven boundaries around LLM tool calls. It intercepts tool executions, evaluates them against declarative policies, and persists their state via customizable Checkpoint Stores and Audit Sinks.

Installation

npm install @agentiam/core

Quick Start

import { createAgentIAM, definePolicy } from "@agentiam/core";

// 1. Define an authorization policy
const policy = definePolicy({
  id: "default-policy",
  rules: [
    {
      id: "require-approval",
      when: { action: "delete_database" },
      decision: "approval_required",
      requirements: ["human_approval"]
    }
  ]
});

// 2. Initialize the IAM engine
const iam = createAgentIAM({ policy });

// 3. Guard a tool execution
const request = {
  actor: { type: "agent", id: "assistant" },
  action: { name: "delete_database", input: { db: "prod" } }
};

const result = await iam.guard(request, async () => {
  return await db.delete("prod");
});

console.log(result.executed); // false (approval required)

Architecture

This package exports the interfaces for CheckpointStore and AuditSink. You can use the provided InMemoryCheckpointStore for local development, or adapt your own databases using these interfaces. See @agentiam/pg for a production-ready Postgres adapter.

For integration with LangGraph, use the @agentiam/langgraph package.

License

MIT