Skip to content

Console

Release binding: v0.1.0-enterprise-ready · AMI ami-010806d4d3445660e · 2026-05-19

The Console is the appliance's built-in UI, served by the gate at:

https://<host>:8443/console

It authenticates with the same credentials as the API and is backed by the read-only /v1/console/* routes plus the live WS /v1/ws/stream telemetry stream.

Screens

Screen What it shows
Command Deck The operational front page: service health, active namespaces, current load
Runtime Engine runtime state and request activity
GPU Runtime GPU utilization, VRAM residency, and serving state
Namespace Cockpit Per-namespace keys, versions, and activity
Representation Inspector A stored entry's representation: cores, rank, fidelity verdict, and compression ratio
Model Certification Per-model retrieval-class membership, cross-linked to the Receipt Inspector for the EXACT badge
Receipt Inspector Paste or upload an ML-DSA-65 receipt and verify it through POST /v1/attestation/verify, with the per-check breakdown inline
Telemetry Per-class query volume and latency tiles backed by /metrics
Flow Trace Request flow tracing across gate and engine
Global Search Search across namespaces, keys, and metadata the credential is authorized for
Settings Console preferences

Notes for operators

  • The Console is served only on TLS 8443, the appliance's single exposed port; restricting the security group restricts the Console with it.
  • Usage and metering views are accounting surfaces, not invoices; see Metering.
  • Alert thresholds surfaced in Telemetry mirror the Retrieval Classes runbook.