The Concierge — Personal AI Access Layer

Unified interface over your entire Tailnet — Hermes Agent as the brain, Tailscale as the identity, every service as a tool

⟫ LAYER 1: IDENTITY (Tailscale MagicDNS + Who-Is-Logged-In headers) Javid's Laptop edminymanizerkilmerlomonsk21@ Phone edminymanizerkilmerlomonsk21@ Tailscale MagicDNS *.chameleon-haddock.ts.net Tailscale-User-* Headers Login, Name, Profile-Pic, ID ↓ Zero Trust ↓ No public ports Tailscale-aware routing Sacred identity inheritance ⟫ LAYER 2: INTERFACES (Choice of front-end — same brain) Telegram Bot Already wired ✓ Home channel: 8590652454 Web UI (HTMX) FastAPI + minimal JS Phase 2 Voice (ElevenLabs) "Hey Concierge..." Phase 3 SMS (Twilio) When no internet Connected ✓ CLI (Hermes REPL) Already exists your mom has it Choice is yours All paths reach the same Concierge API ⟫ LAYER 3: BRAIN (Hermes Agent — reasoning + skills + tools) Concierge API FastAPI + HTMX :8888 (tailnet only) Routes by: • Request type • User identity • Recent context Hermes Agent Reasoning + tool execution Persistent memory Skills: • 200+ skills loaded on demand • Web search, code, vision • Voice, images, video Service Registry "What does Javid have?" Queries live state Knows: • All tailnet services • Their endpoints • Their capabilities Context Engine Per-user memory Recent requests Sources: • Wiki-RAG (970MB) • Hermes memory • Session history Tool Router "Which tool?" "Which service?" → Service call → Wiki search → Glances query → n8n trigger ⟫ LAYER 4: TAILNET SERVICES (existing — all accessible via Tailscale Serve) Glances System metrics CPU, RAM, disk :61208 Homarr Dashboard v1.74.0 :7575 Wiki-RAG Knowledge base SQLite FTS5 970MB Aperture AI gateway 3 providers aperture-hermeshaltsmithey n8n Workflow engine Webhooks bridges Telegram Authentik SSO OIDC modcc-authentik Portainer Container mgmt tailscale-sidecar modcc-tails-portainer Pocket-ID OIDC modcc-pocket-id Audiobookshelf Audio library modccd-audiobookshelf Agent Vault Credential injection modccd-agent-vault + 8 more services All ScaleTail, all tailnet-only glances, homarr, etc. + future services Tool Router adapts Plug in, get discovered wiki search results Legend Interface Brain (Hermes) Tailnet service Knowledge base Identity

Phase 1: Telegram-as-Unified-Interface (Today)

  • • Already wired, just needs skill binding
  • • Add: "what's running?" → service registry query
  • • Add: "ask the wiki" → wiki-rag search
  • • Add: "kill X" → portainer API
  • • Add: "stats" → glances

Phase 2: Web UI (Tomorrow)

  • • FastAPI + HTMX dashboard
  • • Service grid: clickable cards
  • • Command bar at top (HTMX)
  • • Mini Glances widgets inline
  • • Tailscale user identity on every page

Phase 3-4: Voice + Automation (Next Week)

  • • Voice command via ElevenLabs
  • • Per-user adaptive UI
  • • n8n triggers from Concierge
  • • Cross-service rule engine
  • • "Proactive" alerts (anomaly)

The Concierge • Project Level Up Javid • Built on the Tailnet you've already built