Platform components
Cognitora ships six statically linked binaries. Four sit on the inference hot path (router, agent, KV cache daemon, metrics); cgn-ctl is the admin CLI; cgn-operator reconciles Kubernetes CRDs into etcd state the daemons already consume.
Each page follows the same layout as the marketing site platform section: overview, features, architecture, configuration highlights, an example, dependencies, and links deeper into this repo.
| Binary | Role |
|---|---|
cgn-router | OpenAI-compatible HTTP gateway, KV-aware routing, admission, federation |
cgn-agent | Per-node engine supervisor (vLLM / SGLang / llama.cpp / OpenAI-compat proxy) |
cgn-kvcached | Multi-tier KV block store + cross-node QUIC fetch |
cgn-metrics | Prometheus scrape fan-in + power telemetry for routing scores |
cgn-ctl | Cluster ops, PKI, keys, recipes, install/render helpers |
cgn-operator | kube-rs controller for InferenceCluster, ModelPool, RoutingPolicy |
Authoritative TOML schema: configs/cognitora.toml.example and docs/reference/config.md. Source: rust/services/.