aegis-verify

Aegis

Independent verification layer for autonomous AI agent purchases.

Aegis intercepts AI agent purchase intents, runs multi-dimensional verification checks, and returns an approve/flag/block decision — before money moves.

Your AI agent is about to spend $347 on the wrong hotel. Aegis catches that.

Why Aegis?

AI agents are increasingly making autonomous purchases — booking travel, ordering supplies, executing transactions. But there's no standard infrastructure to verify these purchases before they execute.

Agents overpay. They buy the wrong thing. They fall for scam sellers. They agree to hidden subscriptions. Aegis fixes this.

Quick Start

pip install aegis-verify

from aegis import verify_purchase

result = await verify_purchase(
    intent={
        "item": "Sony WH-1000XM5 Headphones",
        "price": 278.00,
        "seller": "electronics-deals-store.com",
        "original_instruction": "best noise canceling headphones under $300"
    }
)

if result.decision == "approve":
    await execute_purchase()
elif result.decision == "flag":
    print(f"Concerns: {result.reasons}")
elif result.decision == "block":
    print(f"Blocked: {result.reasons}")

What Aegis Checks

Integrations

LangChain

from aegis.integrations import LangChainAuditTool
tools = [LangChainAuditTool()]

CrewAI

from aegis.integrations import CrewAIAuditTool
agent = Agent(tools=[CrewAIAuditTool()])

Claude MCP

{
  "mcpServers": {
    "aegis": {
      "command": "aegis-mcp"
    }
  }
}

How It Works

User Request → Agent Proposes Purchase → Aegis Verifies → Approve/Flag/Block → Execute or Halt

See the full Developer Workflow Diagram for details.

Roadmap

• Core verification engine with 5 modules
• LangChain integration
• CrewAI integration
• Claude MCP server
• Hosted API with dashboard
• Seller-side verification tools
• Decentralized validator network

Contributing

Contributions welcome! Open an issue or submit a pull request.

License

MIT — see LICENSE for details.