Show HN: RiskKernel, kill -9 an AI agent and resume it without paying twice

Category: infrastructure

Tags: ai-agents, guardrails, reliability-runtime

Score: 7.8/10 (Innovation: 8, Technical: 8, Documentation: 8, Utility: 7)

RiskKernel is a self-hosted agent reliability runtime that enforces deterministic cost, loop, and time budgets for AI agents, with crash-resumable checkpoints and human-approval gates. It addresses a critical operational gap in production AI agent deployments by providing a SRE-like layer that can be adopted with zero code changes via a proxy or with deeper Python SDK integration. The combination of hard budgets, checkpoint recovery, and OpenTelemetry observability in a single binary is an innovative approach to preventing runaway costs and failures.

Target audience: backend devs, devops, ai-engineers

Repository: https://riskkernel.com/ · Go · Apache-2.0 · 17 stars

View on Hacker News