Das-re

Inspiration

On-call engineers lose valuable time recovering context from past incidents. We wanted to preserve that knowledge across shifts while preventing AI agents from executing destructive fixes.

What it does

Das-re recalls past incidents, maps service dependencies, and screens proposed commands against safety policies. It tests candidate fixes in isolated sandboxes and saves the evidence for future shifts.

How we built it

We built Das-re in Python using AWS Bedrock for reasoning, SurrealDB for memory, Daytona for sandbox execution, and Arize Phoenix for tracing. Anchor Browser supports dashboard reconnaissance, while ElevenLabs provides optional voice alerts.

Challenges we ran into

Coordinating multiple integrations and keeping the demo reliable were major challenges. We also needed to capture incident evidence before sandbox teardown and distinguish successful execution from a safe outcome.

Accomplishments that we're proud of

We connected incident recall, command screening, sandbox verification, and postmortem capture into one workflow. Our Redis lock scenario demonstrates rejecting a destructive command in favor of testing a targeted fix.

What we learned

Safe agents need execution controls alongside observability and evaluation. Incident memory becomes more useful when it preserves service relationships, shift notes, and the evidence behind previous fixes.

What's next for Das-re

We’ll expand incident coverage, strengthen permissions, and improve human approval workflows. Our vision is a future where every shift maintains context and your team can eventually sleep through the outage.

https://drive.google.com/file/d/1k8EaWVOzkg7ei0dmAkURU7-u_xk6f440/view?usp=share_link