Inspiration
As AI agents become autonomous, failures can cause financial losses and erode trust. We wanted to create a system where AI is accountable by design, providing instant, transparent compensation when agents behave unexpectedly.
What it does
AgentIndemnity is a programmable accountability layer for AI agents. It secures agents with USDC-backed performance bonds, detects harmful or policy-violating outputs, and automatically compensates affected users on-chain in real time.
How we built it
- Arc L1 blockchain for low-cost settlement
- Circle Gateway Nanopayments (x402) for per-request payments
- Smart contracts to manage performance bonds and slashing
- AI anomaly detection using Gemini and specialist LLMs
- Middleware SDK for easy developer integration
- Real-time dashboard powered by WebSockets
Challenges we ran into
- Designing accurate AI anomaly detection with minimal false positives.
- Building secure, automated on-chain compensation logic.
- Routing requests efficiently across multiple AI models.
- Keeping transaction costs low while maintaining real-time performance.
Accomplishments that we're proud of
- Built an autonomous AI accountability framework.
- Enabled instant, on-chain USDC compensation.
- Created a flexible multi-model routing architecture.
- Simplified integration through a single middleware layer.
- Delivered live monitoring for developers, compliance teams, and customers.
What we learned
We learned that deploying AI responsibly requires accountability, transparency, and automated enforcement—not just monitoring. Combining blockchain with intelligent routing creates a practical trust layer for autonomous systems.
What's next for AgentIndemnity
- Support additional blockchains and payment networks.
- Expand detection for more AI safety and compliance risks.
- Introduce customizable enterprise policy engines.
- Build integrations for major AI frameworks and cloud platforms.
- Launch a public SDK and marketplace for AI accountability services.
Log in or sign up for Devpost to join the conversation.