Block a user
[Infra] NVIDIA runtime declarative config for Talos
[Infra] Local PV on NVMe for model weight cold starts
[Observability] HPA driven by GPU KV Cache usage
[Observability] Grafana dashboard for TTFT + ITL
[Observability] vLLM Prometheus metrics scraping
[GPU] Preemption swapping for concurrency spikes
[GPU] Enable prefix caching for RAG contexts
[GPU] KV Cache VRAM monitoring + PagedAttention
[Network] Disable proxy buffering for SSE
[Network] HTTP/2 multiplexing for concurrent streams
[Network] TCP backpressure for streaming LLM responses
[Event-Bus E2E] Forgejo PR → event bus → Gotify push + email + observer goal evaluation
[Gateway] Audit trail — log successful JWT auth with identity
[Event-Bus P4] Dead letter queue + Prometheus metrics + Grafana dashboard
[Observer P2] LLM-powered pattern detection + anomaly analysis
[Observer P1] ObserverWorkflow — long-running event evaluator in Temporal
[Event-Bus P3] Target dispatchers — Gotify, SQS, Temporal, webhook
[Event-Bus P2] Source adapters — Forgejo, ArgoCD, Gotify webhooks
[Event-Bus P1] Core event router — POST /v1/events endpoint
[Event-Bus P0] Deploy Gotify notification server