Starling Ops Architecture
Starling is the connectivity layer of the Warble Cloud ecosystem. It implements the Model Context Protocol (MCP) to bridge AI agents to live Kubernetes state without compromising on security or auditability.
System Blueprint
Live Data Flow • JSON-RPC over HTTPS/SSE
Ephemeral Context
Kubeconfigs and secrets are ingested into memory-only buffers. Once the tool call completes, the client is destroyed and memory is flushed. We never write your credentials to disk or persistent logs.
Credit Metering
Every tool call is authenticated and metered via a Firestore-backed ledger. High-intensity tools like scan_cluster cost more than simple lookups, ensuring fair multi-tenant resource allocation.
Mode-Gated Writes
Starling binaries and pods run in readonly mode by default. Destructive actions like patch_deployment are hidden from the LLM unless the server is explicitly launched in read-write mode.
Dual-Transport
One handler, multiple ways to reach it. Use the local stdio transport for your laptop, or the HTTPS/SSE transport for cloud-to-cloud agent communication.
The Tech Substrate
Want to see the product demo?
We've prepared a dedicated slideshow covering the SRE context gap, live remediation flows, and the 3 AM incident response timeline.
Go to Ops Dashboard