play.frakma.io · live demo · GKE warble-byok
The FRAKMA demo stack, running live
This is the Warble Cloud demo cluster: a single-node GKE cluster (warble-byok, us-central1-a) running the retrieval, agent and tracing pieces behind our ML/LLM Ops work. Statuses below are checked live from the cluster. The services stay private; this page only reads their health checks.
Services on warble-byok
Qdrant checking…
Vector database for retrieval. One replica on premium-rwo storage in namespace warblecloud-ai.
reflexion-engine checking…
Actor/critic agent engine that analyzes failures and proposes fixes. A person approves what runs. Backed by Postgres and Redis.
Arize Phoenix checking…
LLM tracing and observability for the agent runs (OTLP on 4317/4318). Namespace phoenix.
Starling EX image pending
Kubernetes workload operator (CRD StarlingWorkload). The CRD and RBAC are installed. The controller image hasn't been published yet.
warble-cli CLI
The command line that plans, dry-runs and checks this stack. --dry-run never applies changes, and the CLI never reads or prints secrets.
How the demo is deployed
gcloud container clusters get-credentials warble-byok --zone us-central1-a kubectl apply -f k8s/namespaces.yaml kubectl apply -k k8s/overlays/warble-byok # Qdrant, reflexion-engine, Phoenix kubectl apply -k k8s/overlays/warble-byok/starling-ex # Starling EX operator
The overlay sizes everything for one e2-standard-2 node: one replica each, with small CPU and memory requests.