AriakiAriaki
I build things that keep running on their own: AI agents, local inference, home infrastructure. Writing the code is half of it; deploying, measuring, and writing it down is the other half.
- public posts
- 13
- writing since
- 2026.02
- main focus
- AI · Infra
What I focus on
AI agents and multi-agent systems
Putting agents into production is a software engineering problem — state, constraints, observability — not a prompting trick.
Local inference and self-hosted hardware
Running models on my own machines, and letting measured numbers decide the GPU, the model size, and what a small model can handle.
Home infrastructure and automation
Smart home, home servers, monitoring, and scheduled jobs — personal infrastructure I maintain, inspect, and change myself.
Writing the conclusions down
An experiment that was never written up did not happen. Every number in a post comes from a run I actually made, including the ones that proved me wrong.
Things I build
The parts you can open.
- 01AlkairaA running simulation where 16 AI characters live autonomously for five years, with a ledger — not narrative — deciding what is true.
- 02Agent Ops DashboardAn ops dashboard for a team of coding agents: non-LLM routing, a kanban board, and context monitoring.
- 03Cyber FireworksA Canvas particle system with trails, glow, and gravity — playable directly in the browser.
- 04smartfridge-mcp-serverAn MCP server implementation with data persistence, an HTTP bridge, and containerization filled in.
Selected writing
If you want a quick read on how I think, start here.
- Narrative Can Lie. The Ledger Cannot: Two Layers of Reality in an AI WorldPart two of the Alkaira series. One AI agent claims to run an underground intelligence network. Another believes she controls his informants. The ledger recognizes neither. On why lying should be allowed, and how a world defends the facts.
- Choosing a Host for an AI Assistant on a $1,399 BudgetFive real options from $249 to $1,399, what each one gives up, and why I ultimately chose something else.
- When AI Agents Reach Production, the Work Is Software EngineeringA prompt is only one component. As an agent moves from demo to production, the hard problems become specification, state management, access control, and failure design—the familiar territory of software engineering.
- Small Models vs. Large Models: What 4B Parameters Can and Cannot DoA real attempt to run a multi-agent system on a local 4B model, from compressing the system prompt to giving up—and the capability boundary that attempt exposed.
Philosophy
I just want things to be the way they should be.
That belief costs rework. The first version of this homepage was a tidy little card, and I rejected it myself — it polished the decoration without answering what a homepage should be. So it was rebuilt as a full-screen point field. I would rather redo the work than ship a version that bothers me.
Contact
A technical question, a collaboration, or just a note — email works.