What it does

Credential Reaper is a AI powered machine identity security platform design to help organize, discover, investigate and manage exposed or potentially compromised credentials.

How we built it

We built Credential reaper using a combination of AI, real time data infrastructure, and a web application stack.

NVIDIA Nemotron: Powers Netrunner, our AI investigation agent. It reasons over available security context, uses application tools to gather information, and generates findings and remediation recommendations.

SpacetimeDB: Serves as our real-time database, maintaining application state and synchronizing security events, investigations, and remediation updates with the frontend.

Cloudflare Workers: Provides the backend API layer and handles communication between the frontend and SpacetimeDB.

React: Powers the user interface, including the investigation dashboard, live security activity feed, remediation workflow, and Afterlife audit history.

GitHub integration: We developed a fake application repository with a bunch of machine identities to provide repository activity for our security workflow. It is connected through GitHub webhooks so repository events can trigger investigations in Credential Reaper. Pull request are also able to be called by our system by the user.

Challenges we ran into

One of our biggest challenges was building an AI agent that could do more than generate reports and explanations, we wanted it to be agentic, and using the tools we would provide. We wanted our agent Netrunner to interact with tools, gather evidence, and produce structured finding that could move an investigation forward.

Another challenge was designing remediation safely. Security automation brings a list of unwanted security risk, if the agent is given too much authority. We had distinguish its role whether proposing a fix, approving proposal, to authorizing actual execution. Each step needed to be reflected accurately in the application's state.

Integrating multiple technology and independently developed components also introduced challenges. Coordinating our GitHub integration, backend changes, real time database state, and frontend remediation workflow required careful integration.

Finally we had to balance a detailed security workflow with interface that was not too cluttered yet still give a specialist the right amount of information to perform their job effectively. That perfect balance of not overwhelming a user or burning them out with endless reading allowed us to come up with precise evidence and decisions without the bloat.

Accomplishments that we're proud of

  • Building an agentic security investigation workflow using NVIDIA Nemotron.
  • Integrating real time application state with SpacetimeDB.
  • Putting humans in control of remediation because we always want a human in the loop.
  • Creating auditable security workflow
  • Building toward real repository driven investigation

What we learned

We learned that applying AI is not just about picking the model or preventing prompt injection. The surrounding system matters just as much, from the quality of evidence, the tools agents can access, the permission it receives, and how its responses are validated

We also learned that the relationship between AI autonomy and human oversight is not inverse. Just because there autonomy there is still the consequential security decision we have to decide.

NVIDIA Nemotron helped us explore different agents that can still support multi step work and reasoning. SpacetimeDB taught us about dynamic/real time DB, and how it can help with events and decisions that are decided by the seconds.

What's next for Credential Reaper

Beyond the hackathon, we envision expanding Credential Reaper into a broader machine identity lifecycle management platform. Future development could include:

  • integration with additional identity providers, cloud platforms, and developer tools.
  • more comprehensive identity ownership, permission and exposure analysis.
  • verified remediation through integrations with credential and access management systems.
  • stronger organization controls, including user authentication, role-based permission, and audit requirement.
  • expanded verification capabilities to confirm that remediation actually resolved the underlying risk.

As organizations continue adopting automation and AI agents, the number of machine identities they depend will continue to grow. We want Credential Reaper to help ensure that those old and forgotten identities are not only discovered, but also understood, governed, and responsibly retired.

Built With

  • cloudflare-workers
  • github-api
  • github-webhook
  • nvidia-nemotron
  • react
  • spacetimedb
  • typescript
Share this project:

Updates

Submission history