> initializing agent.fleet

Building at the agentic frontier

I build the systems that build.

Not web apps — autonomous agents. I build AI that reasons, takes action, and runs itself: a governed fleet that researches, writes code, monitors, and self-heals — 24/7, on a local-first model stack, while I sleep.

Founder @ DCYFR Labs  ·  Head of AI@ GameShark Labs  ·  remote

fleet.stream30+ agents24/7 autonomylocal-first inferencecost-gated
>

I build autonomous AI systems that operate the same tools people do.

They reason, take action, and run entire workflows — research, code, ops — for hours or months, so I ship outcomes, not keystrokes.

Chat answers. Agents act. I operate the fleet.

Selected work

Systems that run themselves.

Not a stack of frameworks — the autonomous machinery I've built, shipped, and keep running.

01

Autonomous Agent Fleet

A governed swarm of AI daemons — research, code review, monitoring, self-healing — running a 30-minute cognitive loop on a local-first model stack, under a hard spend gate with metacognition to stop it spinning.

Multi-agentLocal-first inferenceSelf-healingMetacognition
Running
02

DCYFR AI

The portable TypeScript runtime my agents think in — model routing across local and frontier tiers, tool-use, and MCP wiring you drop into any project.

Agent runtimeTool-useMCPTypeScript
Open
03

Agent Governance & Safety

The rails that let agents take real actions without going off the rails — sandboxes, kill-switches, per-process credential isolation, and enforced spend limits. Autonomy you can trust.

AI safetySandboxingKill-switches
Live
04

SharkVault

Proof the fleet ships real product: a backer-funded consumer app taken end-to-end — auth, payments, content pipeline — largely by agents.

Autonomous deliveryProductNext.js
Live

Ventures

What I lead

Writing & garden

Field notes from the frontier.

Essays when they're finished; a digital garden of evergreen notes while they're still growing.

Now · last synced Jul 2026

Current focus

Teaching the fleet to plan. Multi-step missions the agents decompose, execute, and grade themselves — with me out of the loop.

Raising the trust ceiling. Hardening the safety model so agents can take higher-stakes actions without a human gate on every step.

Let's build something autonomous.