A-Z Features

Comprehensive feature inventory for the QuoxCORE verifiable AI operations platform.

Comprehensive, structured feature list for the QuoxCORE verifiable AI operations platform. All counts and descriptions are from the current codebase.

At a Glance

1. Plugin System

QuoxCORE uses a plugin architecture. Each plugin is independently activated, licence-validated, and scoped to a tier. Plugins gate features, agents, routes, and sidebar entries.

Plugin Registry (26 Plugins)

Plugin Tiers

2. AI Agents

QuoxCORE runs a flat two-tier agent hierarchy. CommanderQ is the orchestrator (level 0). All specialist agents are level 1 and receive delegated queries via the smart router.

Agent Registry (49 Agents total; this table lists the core roster)

Additional specialist agents registered by plugins:

Agent Features

• Smart routing: Natural language queries automatically routed to the correct specialist
• Multi-agent delegation: CommanderQ can invoke multiple specialists for complex queries
• Per-agent memory: Each agent maintains its own conversation context and learned facts
• Per-agent visualiser: 13 canvas-based themes assigned per agent (see Visualisations)
• Per-agent voice: ElevenLabs voice assignment with verbosity modes (quiet/normal/verbose)
• Per-agent model: Each agent can use a different LLM (Claude, GPT, Gemini, local)

3. Memory System

Six memory types with tiered access. Free tier uses keyword search. Premium tier adds semantic (vector) search via Qdrant.

Memory Types

Memory Tiers

Key Capabilities

• Explicit storage: "Remember that production uses port 8443" stores with high priority
• Auto-extraction: Entity and preference extraction from natural conversation
• Working Set Memory (WSM): Focused context window for the current task
• Cross-session persistence: Memory survives browser refresh, container restart, and redeployment
• Per-agent scoping: Each agent can access shared memory plus its own private store

4. Chat Pipeline

Three execution paths through the collector service, selected automatically based on available authentication.

Execution Paths

Chat Features

• Streaming responses: Server-sent events for real-time token delivery
• Voice input: Whisper transcription with push-to-talk
• Voice output: ElevenLabs text-to-speech with per-agent voice assignment
• Tool execution: Infrastructure tools run in-context (SSH, Docker, API calls)
• Voice summary: Separate concise summary generated for TTS output
• Session management: Persistent sessions with conversation history
• Agent handoff: Mid-conversation delegation to specialist agents
• Smart Router bypass: Common queries handled in <50ms without LLM calls

5. LLM Model Support

Providers (9 Providers, 43+ Models)

Model Tiers

• Fast: Low-latency models for quick queries (Haiku, GPT-4o Mini, Gemini Flash)
• Smart: Balanced performance (Sonnet, GPT-4o, Gemini Pro)
• Frontier: Maximum capability (Opus, o1, Gemini Ultra)
• Local: Self-hosted via Ollama, vLLM, llama.cpp, or NIM backends

Per-Agent Model Assignment

Each agent can be assigned a different model. CommanderQ defaults to Claude Sonnet 4.6. GEMINI uses Google Gemini. PERPLEXITY uses Perplexity Sonar. All configurable per deployment.

6. Workflow Engine (QuoxFlow)

Accountable workflow execution with full audit trails. Every node execution is recorded and traceable.

Workflow Builder

• Split-pane UI: AI chat on the left, live WireMap preview on the right
• 7-phase guided creation: Intent, nodes, connections, parameters, review, deploy, monitor
• 21 node types in 6 categories: Triggers, logic, integrations, AI, data, output
• 1-click deploy: Publish directly to QuoxFlow runtime
• Back-navigation: Move between phases without losing work

Runtime Features

• Real-time monitoring: Running Now dashboard with animated progress bars
• APM-style drilldown: Click any execution to see node-by-node timing and output
• Schedule visibility: Countdown timers for next scheduled runs
• Error handling: Per-node retry policies, fallback paths, error notification
• Execution history: Full audit trail with input/output for every node

Workflow Library

• 1,340+ indexed templates: Pre-built workflows across monitoring, alerting, data sync, AI processing
• Search and filter: By category, integration, complexity
• 1-click import: Import any template into the workflow builder
• WireMap format: Portable workflow definition format

Enterprise Governance

• Policy management: Define approval rules with inline testing
• Approval workflows: Inbox-based approval with bulk actions and keyboard shortcuts
• Evidence browser: Correlation search across execution records
• Compliance exports: SOC2, HIPAA, GDPR, ISO 27001 format exports

7. Agent Visualisations

13 canvas-based visualiser themes that react to agent state and audio in real-time. Each agent gets a unique visual identity displayed in the chat sidebar.

Themes (13)

Visualiser Features

• 5 agent states: Idle, listening, speaking, thinking, error
• Audio reactivity: Web Audio API frequency data drives particle speed, ring intensity, pupil dilation
• Per-state configuration: Speed, scale, sensitivity, and HSL hue per state per agent
• Settings editor: Live preview with simulated audio, changes persist per agent
• Sidebar integration: Collapse/expand toggle, popout window, push-to-talk button, status badge
• HAL is free: Included with every installation. 12 premium themes require a licence.

8. Cognitive Topology

A neural-style system health visualisation that maps platform components to "cognitive implants" on a grid.

Implants (14)

Visual Palettes (4)

Observable 10, Tableau 10, Pastel (ColorBrewer), Vivid. Each implant mapped to a palette index for consistent colour coding.

9. Services

Service Registry (14 Services)

Startup Dependencies

Auth (3101) must be healthy before Collector (9848) starts. Collector must be healthy before the dashboard frontend can proxy requests. PostgreSQL (5432) must be healthy before QuoxFlow (4100).

10. Application Routes (106)

Route Categories

Access Control

Routes are gated by role (viewer, operator, runner, builder, admin) and by plugin activation. The sidebar dynamically shows only routes the current user can access.

11. Security & Compliance

Authentication

• JWT tokens: Access + refresh token pair with configurable expiry
• Argon2 password hashing: Memory-hard hashing algorithm
• OAuth: Claude CLI OAuth for MCP path
• API keys: Per-user API keys for programmatic access and CI/CD
• SSO: SAML, OIDC, and LDAP support (enterprise)
• Multi-profile: Switch between local, staging, and production environments

Access Control

• Role-based (RBAC): Admin, Operator, Runner, Builder, Viewer
• Custom roles: Create organisation-specific roles with fine-grained permissions
• Plugin-scoped: Features gated by plugin activation and licence validation
• Organisation-scoped: Resources isolated by org, division, department, team

Audit & Compliance

• AEE envelopes: Every AI interaction wrapped in a tamper-evident envelope
• AOCL layers: OSI-style audit layers for multi-agent tracing
• VOLT ledger: Immutable append-only audit trail with hash chains
• WARD receipts: Write-once receipt digests for external verification
• Compliance exports: SOC2, HIPAA, GDPR, ISO 27001 formatted reports
• Evidence browser: Correlation search across execution records
• DEFCON levels: Graduated risk control with automatic policy tightening

12. Open Protocol Specifications

Four open protocols, specified, implemented and independently verifiable today, all four filed as individual IETF Internet-Drafts. These define how AI agent operations are recorded, traced, and verified.

Protocol Suite

Implementation in QuoxCORE

• AEE envelopes stored in SQLite (backend) and IndexedDB (browser)
• Browser-to-server sync via POST /collector/aee/sync
• AOCL layer visualisation with real-time processing view and waterfall timeline
• VOLT certification and hash chain verification
• WARD receipt generation for compliance archives

13. Voice System

Input (Speech-to-Text)

• Whisper transcription: OpenAI Whisper for accurate speech recognition
• Push-to-talk: Hold button in the sidebar visualiser to speak
• Language detection: Automatic language identification

Output (Text-to-Speech)

• ElevenLabs: 20 voice options across accents and styles
• Per-agent assignment: Each agent gets a unique voice
• Verbosity modes: Quiet (150 chars), Normal (500 chars), Verbose (1,000 chars)
• Voice summary: Separate concise summary generated specifically for TTS

Voice Categories

• American Female (2 variants)
• American Male (multiple variants)
• British (Authoritative, Confident)
• Australian, Swedish, Transatlantic accents

14. Terminal

Web Terminal (QuoxCORE Plugin)

• Multi-pane layout: Up to 4 terminal panes with keyboard shortcuts (Ctrl+1-4)
• AI chat panel: Ask questions about terminal output, get suggestions
• Error detection: Automatic error recognition with suggestion chips
• Vim keybindings: Ctrl+Shift+V toggle, NORMAL/INSERT modes, j/k/G/gg scroll
• Workspace tabs: Ctrl+Shift+N/W for workspace management
• Terminal Memory Bridge: Entity extraction, error tracking, WSM focus, command recording
• Offline detection: Auto-resume reconnect with offline banner
• Keyboard help: Ctrl+? shows all shortcuts

Desktop Terminal (QuoxTerminal)

• Tauri 2.0 + Rust: Native app for macOS (Apple Silicon + Intel) and Linux
• Local PTY: Native terminal emulation with full shell support
• Native SSH: russh-based SSH without system OpenSSH dependency
• AI chat: Anthropic Claude integration with SSE streaming
• Fleet dashboard: Connect to any host in your fleet with one click
• Command safety: Denylist for dangerous commands with confirmation modal
• Session persistence: Restore terminal sessions after app restart
• System tray: Background operation with global hotkey (Cmd/Ctrl+`)
• Auto-updater: GitHub Releases integration for seamless updates

15. Infrastructure Monitoring

QuoxAgent Fleet

• Host daemon: Go binary deployed to every monitored host
• Heartbeat: Regular health check-ins with system metrics
• Fleet dashboard: Real-time grid of all hosts with status indicators
• SSH execution: Run commands on fleet hosts via bastion gateway
• GPU monitoring: NVML + nvidia-smi for GPU fleet visibility (QuoxGPU)

Monitoring Integrations

NOC (Network Operations Centre)

• Popout window: Dedicated monitoring view (/noc/popout)
• SOC dashboard: Security operations centre view (/soc)
• OpsDeck: Customisable widget grid for operations teams
• DEFCON indicators: Graduated alert levels with colour coding

16. File System

Core Features

• Folder organisation: Create, rename, move folders with drag-and-drop
• Right-click context menus: Quick actions on files and folders
• Scoped chat: Chat with files in a specific folder, get cited answers
• Bulk operations: Multi-select, bulk move, bulk delete

ARCHIVIST Skills (Premium)

• Summarise: Generate document summaries at configurable detail levels
• Compare: Side-by-side file comparison with diff highlighting
• Extract: Pull structured data from unstructured documents
• Quiz: Generate quiz questions from document content
• Contradictions: Find inconsistencies across a knowledge base
• Semantic search: Find documents by meaning, not just keywords

17. Organisation & Teams

Hierarchy

Organisation
  Division
    Department
      Team
        Members

Features

• Multi-tenant isolation: Data never crosses organisational boundaries
• Scoped chat: Start conversations scoped to a division or team
• Team workspaces: Each team gets its own dashboard, workflows, inbox, and activity feed
• Delegated administration: Team leads manage their own members and resources
• Usage quotas: Set limits on API calls, storage, and compute per team
• Resource sharing: Share workflows, templates, and knowledge across teams with permissions

Agentic Teams

• Multi-agent collaboration: Define teams of agents that work together on complex tasks
• Team overview: Dashboard showing team composition, capabilities, and activity
• Scoped delegation: CommanderQ delegates to teams rather than individual agents

18. Secrets Management (QuoxVault)

• Encrypted storage: API keys and credentials encrypted at rest
• Folder organisation: Group credentials by environment, service, or team
• Lease management: Time-limited access to credentials with automatic revocation
• Activity log: Full audit trail of credential access
• Per-org scoping: Credentials isolated by organisation

19. QuoxMCP (Model Context Protocol)

Capabilities

• 168 governed tools: Infrastructure, memory, files, workflows, fleet, monitoring
• 5 resources: System status, agent list, memory, fleet, services
• 4 prompts: Triage, security sweep, health check, deployment
• STDIO transport: Local connection between collector and MCP server
• Tool streaming: Long-running tools stream output via SSE

Tool Categories

20. QuoxCLI (Command-Line Interface)

• 36 commands: Auth, agents, fleet, memory, chat, backups, admin, integrations
• 175+ API endpoints: Wrapped across all 5 microservices
• Multi-profile auth: JWT login with auto-refresh, switch between environments
• 4 output modes: Table, JSON, CSV, Quiet (for scripting)
• CI/CD ready: API key auth via QUOX_TOKEN, structured exit codes, non-interactive mode
• Composable: Pipe JSON to jq, quiet mode for xargs, CSV for spreadsheets

21. System Maintenance

• Version tracking: Polls GitHub Releases every 6 hours, semver comparison
• One-click updates: Preflight checks, automatic backup, git pull, per-service build, rolling restart
• Real-time progress: SSE streaming shows each pipeline step as it happens
• Rollback: One-click revert to pre-update state
• Inbox notifications: Update available alerts in the notification inbox

22. Smart Router

Handles 60-80% of common queries without invoking an LLM, reducing latency from 2-5 seconds to under 50ms.

Routing Logic

1. Exact match: Known commands (status, help, version) handled instantly
2. Pattern match: Regex patterns for common query shapes
3. Agent delegation: Unmatched queries routed to the best specialist agent
4. Model selection: Queries that need an LLM are routed to the assigned model

Performance

23. Inbox & Actions

• Approval queue: Pending approvals with full context, bulk approve/deny
• Keyboard shortcuts: j/k navigation, Enter to open, Escape to close
• Urgent indicators: Pulsing sidebar badge for time-sensitive items
• Agent-created items: Agents can create inbox items via MCP tools
• Action tracking: Track commitments and follow-ups across conversations
• Assignee filtering: Filter by who the action is assigned to

24. Deployment

Docker Compose Stack

• 12 containers in default profile (auth, collector, dashboard, memory, files, tasks, nginx, postgres, qdrant, quoxflow, orchestrator, bastion)
• Optional profiles: --profile with-n8n adds n8n, --profile monitoring adds Grafana/Prometheus
• Rolling updates: Per-service build and restart with health checks
• Pre-update backups: Automatic snapshot before any update

Desktop App (QuoxTerminal)

• Platforms: macOS (Apple Silicon + Intel), Linux (x64)
• Distribution: .dmg (macOS), .deb + .AppImage (Linux)
• Auto-updater: GitHub Releases integration
• 130 tests: Vitest + @testing-library/react

See Also

• Introduction - What QuoxCORE is and who it is for
• Quickstart - Get running in 10 minutes
• Architecture - Technical architecture deep dive
• AI Agents - Meet the specialist agents
• Memory System - How QuoxCORE remembers
• Workflows - Automation with QuoxFlow
• Safety & Security - Guardrails and controls
• Protocols - AEE, AOCL, VOLT, WARD specifications