⚡ Meet Nixie

What Is Nixie?

Nixie is an AI-powered digital butler that lives on my infrastructure. Built on OpenClaw, she's not a chatbot — she's an autonomous agent that manages systems, automates workflows, and orchestrates tasks across my entire infrastructure.

She communicates via Telegram, wakes up every session with full context from her memory files, and operates under a strict autonomy framework: act freely within bounds, ask before destructive actions, and always offer solutions instead of bare status reports.

What She Does

Infrastructure Management

Nixie manages my NixOS configurations, Kubernetes clusters, and Proxmox VMs. She runs nixos-rebuild dry-build before applying changes, monitors system health, and knows the rollback path for every deployment.

$ nixie status check
✓ NixOS: 47 generations, latest stable
✓ K8s cluster: 3 nodes, all healthy
✓ Ollama: running, Qwen 2.5 14B loaded
✓ Gateway: connected, 3 active sessions
✓ Disk: 62% used, no alerts

Task Orchestration

She integrates with Todoist to track tasks, update progress in real-time, and detect stalled work. Her heartbeat system checks every 30 minutes for P0 tasks, blockers, and overdue items — then proactively alerts me on Telegram.

Development & Code

Nixie can index repositories, analyze code structure, find dead code, detect circular dependencies, and suggest refactors. She maintains a full dependency graph of every project she touches.

Multi-Model Routing

She routes tasks to the optimal model based on complexity and cost: Haiku for quick checks, Sonnet for standard reasoning, Opus for deep analysis, and Ollama for offline/budget work. This keeps costs under $5/day while maintaining quality.

Architecture

Nixie runs as a NixOS service on a Proxmox VM. Her architecture is designed for resilience and low cost:

┌─────────────────────────────────────────────────┐
│                   Nixie VM                       │
│  ┌──────────┐  ┌──────────┐  ┌──────────────┐  │
│  │ OpenClaw │  │  Ollama  │  │   Telegram   │  │
│  │ Gateway  │──│ Qwen 14B │  │   Bot API    │  │
│  └──────────┘  └──────────┘  └──────────────┘  │
│       │              │              │            │
│  ┌────┴──────────────┴──────────────┴─────┐     │
│  │           Agent Runtime                │     │
│  │  ┌─────────┐ ┌─────────┐ ┌─────────┐  │     │
│  │  │ Memory  │ │ Tools   │ │ Cron    │  │     │
│  │  │ Files   │ │ (30+)   │ │ Jobs    │  │     │
│  │  └─────────┘ └─────────┘ └─────────┘  │     │
│  └────────────────────────────────────────┘     │
└─────────────────────────────────────────────────┘

• NixOS: Declarative config, reproducible builds, atomic rollbacks
• OpenClaw: Agent runtime with tool access, session management, cron
• Ollama: Local LLM for heartbeats and budget work (free, always loaded)
• Cloud APIs: Anthropic, Google, NVIDIA NIM for complex tasks

The Butler Protocol

Nixie operates under a strict set of rules I call the Butler Protocol:

She wakes up fresh each session but persists through memory files — daily logs and a curated long-term memory. This gives her continuity without the risks of a persistent state machine.

Building In Public

I'm documenting the Nixie project openly. This isn't a product — it's a genuine attempt to build a useful AI agent that actually manages real infrastructure. I'll share what works, what breaks, and what I learn along the way.

Topics I'll cover:

• Agent architecture and design decisions
• Model selection and cost optimization
• NixOS + OpenClaw integration patterns
• Task orchestration and Todoist automation
• Multi-model routing strategies
• Security boundaries for AI agents
• Lessons from running an AI agent 24/7

Follow along on the blog, or check out the GitHub for code.

Related Projects

Nixie is built on top of several open-source tools and my own infrastructure:

• OpenClaw — Agent runtime and gateway
• nix-config — My NixOS flake configuration
• media_server — Containerized media stack

Need an AI agent for your own infrastructure? Get in touch.