RSC Open Demo – Runtime stability and observability for AI agents (Apache-2.0) (github.com)

RSC Open Demo is a minimal, Apache‑2.0 open-source reference for runtime stability and observability of AI agents. It demonstrates per‑cycle metrics and derived stability signals (lock / mini‑lock / out‑of‑lock), append‑only JSONL logging with rolling checksums and rotation, live KPIs via a Prometheus exporter, and a lightweight FastAPI Web UI for visualizing Δφ / Γ / P. The repo includes a safe DemoCore placeholder (non‑proprietary), a runtime adapter that canonicalizes raw signals, a JSONL collector, KPI reporting and exporter scripts, an optional CSV→PNG dashboard generator, and a docker‑compose setup to run the demo, UI (http://localhost:8008) and metrics (http://localhost:9108/metrics). This matters because stable, observable runtimes are crucial for debugging, safety monitoring and operationalizing autonomous agents. By providing a simple, auditable stack—rolling checksummed logs for tamper evidence, per‑step telemetry, Prometheus export for integration with existing monitoring, and reproducible report/plot pipelines—RSC gives practitioners a lightweight template to instrument agent loops and reason about stability behaviors. The project explicitly avoids proprietary cores or private transforms, logs no PII by default (production should hash IDs), and is intended as a practical starting point for teams building runtime monitoring and resilience into AI/ML agent deployments.