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.

BinaryRole
cgn-routerOpenAI-compatible HTTP gateway, KV-aware routing, admission, federation
cgn-agentPer-node engine supervisor (vLLM / SGLang / llama.cpp / OpenAI-compat proxy)
cgn-kvcachedMulti-tier KV block store + cross-node QUIC fetch
cgn-metricsPrometheus scrape fan-in + power telemetry for routing scores
cgn-ctlCluster ops, PKI, keys, recipes, install/render helpers
cgn-operatorkube-rs controller for InferenceCluster, ModelPool, RoutingPolicy

Authoritative TOML schema: configs/cognitora.toml.example and docs/reference/config.md. Source: rust/services/.

See also