Skip to content

Admin Backend – Overview

The Admin Backend (http://localhost:8088) is the central control panel for the Sovereign MoE cluster. It is accessible to administrators only.

Login

Admin Login

Administrators can authenticate via local credentials or via SSO (Authentik). The login form supports CSRF protection.

Dashboard

Admin Dashboard

The navbar groups all sections into dropdown menus, visible from ≥992 px screen width. On smaller screens a hamburger menu opens the full offcanvas sidebar.

Menu Item Path Function
Config / System configuration, container status

Monitoring (dropdown)

Menu Item Path Function
Monitoring /monitoring Prometheus metrics, server status
Live Monitoring /live-monitoring Active processes, process kill, LLM instances
Starfleet /starfleet Ambient intelligence dashboard, node health, alerts
Pipeline Log /pipeline-log Per-request routing decisions, filters, CSV export
Statistics /statistics Token and cost statistics
Benchmarks /benchmarks GAIA / LongMemEval benchmark runs

Infra (dropdown)

Menu Item Path Function
Servers /servers Configure inference servers
Knowledge /knowledge Neo4j knowledge graph management
Federation /federation Multi-tenant cluster federation
Quarantine /quarantine Blocked content review
Maintenance /maintenance System maintenance tasks

Enterprise Stack (dropdown)

These pages integrate the optional Enterprise Stack (Marquez · lakeFS · NiFi). They appear when the stack is reachable and surface a Foundry-equivalent catalog/approval/explorer/notebook workflow without leaving the admin UI.

Menu Item Path Function
Enterprise Dashboard /enterprise Lineage runs, lakeFS commits, NiFi ETL submissions, drift events
Data Catalog /catalog Cross-source dataset browser (Marquez · Neo4j domains · lakeFS repos)
Knowledge Approvals /approval Review and approve/reject pending knowledge bundles staged on pending/* lakeFS branches
Object Explorer /explorer Read-only Cypher console with regex blacklist + Neo4j Browser deep-link
Notebook /notebook Embedded JupyterLite + copy-paste snippets for the orchestrator API

Tools (dropdown)

Menu Item Path Function
CC Profiles /profiles Claude Code profiles
Skills /skills Manage slash commands
MCP Tools /mcp-tools Enable/disable precision tools
Tool Eval /tool-eval Tool usage evaluation log
Templates /templates Manage expert configurations

Users (dropdown)

Menu Item Path Function
Users /users User CRUD, budgets, permissions, API keys
Teams /teams Team and tenant management
User Content /user-content All user templates and profiles

Dashboard – Global Configuration

The Admin Dashboard (/) allows changing system-wide settings. Changes are written to the .env file via POST /save.

Configurable Fields

Field Env Variable Meaning
Token Price TOKEN_PRICE_EUR EUR per token for cost calculation (default: 0.00002)
Expert Models EXPERT_MODELS JSON array: model assignments per category
Inference Servers INFERENCE_SERVERS JSON array: server configurations
Claude Code URL CC_API_BASE Base URL for Claude Code API
Planner Model PLANNER_MODEL Global default planner LLM
Judge Model JUDGE_MODEL Global default judge LLM
SMTP SMTP_HOST etc. Email configuration for user notifications

Container Status

The dashboard shows the live status of these Docker containers:

  • langgraph-orchestrator
  • moe-kafka
  • neo4j-knowledge
  • mcp-precision
  • terra_cache (Valkey)
  • chromadb-vector

Inference Servers

Inference Servers

Skills

Skills Management

Tool Evaluation Log

Tool Eval

Cluster Impact

Every change made via the Admin Backend takes effect immediately:

Action Immediate Effect
Change token price Applies immediately to all new requests
Change inference servers New requests use the updated server list
Toggle profile Orchestrator is restarted
Suspend user Valkey cache is immediately invalidated
Revoke permission Next API request from the user will be rejected
Set budget Valkey counter is validated against the new limit