AgentLiability

• 

  AgentLiability Dashboard: Setting AI spending guardrails and automated liability settlement

Inspiration

Autonomous AI agents are buying things on our behalf, but what happens when an agent orders the wrong item, gets overcharged, or buys something completely off-policy? We realized there was no safety layer between an AI making a purchase decision and real money leaving an account. That inspired us to build AgentLiability—a spending guardrail and liability settlement protocol for AI agents.

What it does

AgentLiability sits between AI agents and PayPal transactions. It signs preset spending limits, holds funds securely via PayPal, evaluates transactions against strict policies before money is released, and maintains an immutable audit trail of every verdict. If an agent buys the wrong item, the system automatically flags the dispute, freezes funds, and settles liability according to policy.

How we built it

• Frontend: Built with React, Vite, and Tailwind CSS to create a clean, real-time dashboard and transaction console.
• Backend: Node.js & Express API hosted on Render.
• AI & Reasoning: Integrated Groq and Gemini APIs to evaluate agent purchase intents and detect policy violations on the fly.
• Payments: Integrated PayPal SDK / Sandbox APIs for secure pre-authorizations, holds, and automated settlement workflows.

Challenges we ran into

• Webhook & Sandbox Triggers: Handling PayPal sandbox webhook events smoothly without dropping state during local and production builds.
• Context Preservation: Ensuring the AI evaluator received enough contextual metadata to distinguish between legitimate purchases and policy violations without causing high latency.
• Deployment Pipeline: Configuring monorepo root paths correctly across Vercel for the frontend and Render for the backend.

Accomplishments that we're proud of

• Successfully connecting an AI policy evaluator directly into live PayPal transaction flows.
• Delivering a full-stack, end-to-end working MVP with real-time risk assessment and automated dispute logging.
• Creating a clean, intuitive UI that makes complex AI transaction logs and liability disputes easy to understand.

What we learned

• How to manage pre-authorizations and complex transaction states using PayPal APIs.
• Practical techniques for prompt engineering and structured JSON outputs from Gemini and Groq for reliable decision-making.
• Deploying and configuring decoupled full-stack monorepos under tight competition deadlines.

What's next for AgentLiability

• Multi-Agent Governance: Adding multi-signature approvals for high-value AI agent purchases.
• Mainnet PayPal Integration: Moving from sandbox testing to production live credentials with dispute webhooks.
• Custom Policy Builders: Allowing users to define complex visual policy logic for agent spending limits.

Built With

• express.js
• google-gemini
• groq
• mongodb
• node.js
• paypal
• react
• render
• tailwindcss
• vercel
• vite