$ uptime --live · updated September 8, 2026
TheStack.NoSmoke.
Models
Claude Opus 5 + Fable 5/5.1 lead; 24+ models routed by task type
Claude Opus 5
Claude (first-party subscription)
Heavy reasoning — deep development at the desk via Claude Code
Fable 5.1
Claude (first-party subscription)
Frontier coding — deep dev at the desk + remote control from the phone
Fable 5
Claude (first-party subscription)
Agentic coding + client app builds (Archive Salon, beta loop) — desk and remote
GLM 5.3 Flash
Ollama Cloud
Hermes default — desk conversations + Telegram on the phone + cron fleet
GLM 5.3
Ollama Cloud
FCC proxy Opus-tier mapping + heavy cloud coding
Kimi K2.7 Code
Ollama Cloud
Coding sub-agents + delegate tasks
DeepSeek V4 Flash
Ollama Cloud
Fallback — coding + robust tasks
Gemma 4
Ollama Cloud
Vision + image analysis + goal-mode judge
Minimax M3
Ollama Cloud
Content drafting + creative tasks
Granite 4.1
Ollama Local
Local — offline + emergency
Services
All running on a single WSL2 box, 127.0.0.1 only
Hermes Gateway
:8642AI orchestrator + Telegram bot
Python, systemd
FCC Proxy
:8082Claude Code CLI → subscription + cloud model routing
Python, uvicorn
Ollama
:11434Local + cloud model inference (20+ models)
Go, systemd
MemPalace
Persistent memory + knowledge graph
Python, MCP
GitNexus
Code knowledge graph (68K+ nodes indexed)
Node.js, CLI
Open WebUI
:3000Browser-based chat interface
Docker
Open Design
:4000Design AI stack + daemon
Node.js, systemd
Penpot
:9001Self-hosted design platform (Figma alt)
Docker
Postiz
:4007Social media scheduling (14 platforms)
Docker
Toolchain
What each layer does
Orchestration
- ▸Hermes Agent v0.20.2
- ▸delegate_task subagents
- ▸Loop + Goal Mode
- ▸45 cron jobs
Coding
- ▸Claude Code (Opus 5 + Fable 5/5.1 first-party)
- ▸GitNexus code graph
- ▸Agent-Skills (24 SDLC skills)
- ▸Cline CLI
Client Ops
- ▸jts-client-kit (request loop + triage cron)
- ▸Throwaway-DB test isolation
- ▸Gated deploys w/ health checks
- ▸Per-client skill library
Media & Design
- ▸Remotion storyboard video pipeline
- ▸Mermaid diagram-as-code
- ▸Headless-Chrome asset capture
- ▸shotcraft motion library
- ▸Penpot + Open Design
Memory
- ▸MemPalace MCP
- ▸Semantic search
- ▸Knowledge graph
- ▸Session diary
Models
- ▸Claude Opus 5 + Fable 5/5.1 (first-party)
- ▸Ollama Cloud (primary)
- ▸8-model fallback chain
- ▸Granite 4.1 (local)
Infrastructure
- ▸WSL2 Ubuntu
- ▸Systemd services
- ▸19 Docker containers
- ▸iptables firewall
- ▸Client VPS fleet (Hostinger)
How It Actually Works
At the desk: Hermes (the orchestrator) runs conversations and the cron fleet on GLM 5.3 Flash, and deep development goes through Claude Code on frontier models — Opus 5 for the hardest reasoning, Fable 5.1 for the build-and-ship pipeline. Away from the desk: the same stack follows — Hermes answers on Telegram from the phone, and Claude Code keeps building via remote control. MemPalace remembers everything across sessions and surfaces. GitNexus maps the codebase so changes are surgical, not guesswork. Flat subscriptions — no per-token costs, at the desk or in your pocket.
Config auto-backed up weekly to a private GitHub repo · Stack data updated: September 8, 2026