Product

Kan — the eye in the cloud.

"Kan" is கண் — eye in Tamil. The Kan control plane is a multi-tenant SaaS that watches every Hux you've enrolled. It's where you see your fleets, manage your keys, set your budgets, audit what happened, and never look at a YAML file again.

What Kan gives you.

One dashboard, every fleet. Your laptop demo, your dev cluster, your prod region — all in one console. Switch between them in a dropdown.
Keys + groups + budgets. Issue API keys, group them by customer or workload, set rate limits and token / USD budgets per period. Every change propagates to every Hux in 15 seconds.
Audit + observability. See every config change, every key issued, every revocation. See live GPU utilization, request rate, p95/p99 latency, error breakdown. Drill into any model, any node, any key.
Model lifecycle. Adopt vLLM containers your team already runs, or deploy new ones from the Models page. Add replicas across nodes for redundancy. Drain and decommission gracefully.
Zero-config TLS. Each Hux generates a self-signed cert on first boot. Want a real cert? Upload one in the Certificates page and Kan ships it to every Hux in your fleet.

Multi-tenant by construction.

Every row in our Postgres has an org_id and Postgres-level row-level security policies enforce isolation — no application-level bug can leak data across tenants. The fleet enroll token an agent uses scopes its access to its own org.

What Kan never sees.

Prompts, completions, model weights, custom checkpoints. Kan only ingests metering events from your fleets: request_id, timestamp, model name, status code, latencies, token counts, hashed client IP, error code. The data path between your end user and your GPU never crosses our wire.