[ ONLINE ] fleet · self-hosted · guardrailed

root@fleet:
autonomous ai
— self-hosted, no leash.

20 years in the trenches of security. Now I run a full AI fleet on budget Intel Arc silicon — local models, autonomous agents, a memory brain. All of it mine. The useful pieces ship to GitHub.

gpu arc·b60models localagents activeguardrails onuptime 99.9%
Systems

see the fleet, not just the words.

// neuralis · live recall graph · drag to rotate
Memory · Core

Neuralis

A persistent brain that remembers across every tool and agent. Every node is a memory; every edge is a link it made on its own — hybrid semantic + graph recall, self-correcting. The spine the whole fleet runs on.

hybrid recallgraph linkedself-correcting
Open live demo ↗
AURORA-XLIVE
6agents
$0.00spend/hr
99.9%uptime
"summarize today's fleet activity" → done
Control

AURORA-X

One glass pane over the entire fleet — voice, vision, live status, cost, and a chat that drives real work. Twelve skins, mobile-ready, wired to every agent in the stack.

voice + visionlive control12 skins
Open live demo ↗
sentinel-prime · autonomous run

        
Agents

Sentinel Prime

Autonomous desktop & fleet agents. Hand it a goal; it plans, executes, and reports — with guardrails clamped on money and anything outbound. Local-first, so most runs cost $0.

autonomousguardrailedlocal-first
Watch it run in AURORA-X ↗
How it works

the brain stays in the loop.

Every agent run rides the same closed cycle — pull what's known, act inside guardrails, write the result back. That's why the fleet gets sharper the more it runs.

01
Recall
surface what's known
The brain returns relevant memory, context, and past decisions before a single step runs.
02
Act
plan & execute
Agents plan the work and run it locally — money and outbound actions clamped by default.
03
Remember
write it back
Outcomes, fixes, and facts persist to the brain — so the next run starts one step ahead.
a closed loop — every run makes the next one smarter.
The Stack

and the tools that keep it sharp.

Coding
Maestro

Guardrails, memory & cost telemetry bolted onto coding agents so they stay safe and cheap.

Inference
Intel Arc Stack

Local LLM serving on Arc B60 / Battlemage — the tuning & benchmarks that make budget silicon run models.

Ethos
Open Source

The useful pieces are public. Ship it, measure it, publish the numbers, keep innovating.

Featured Work

git remote: open-source & battle-tested.

Principles

non-negotiables.

Self-hosted
Runs on my own silicon — no rented dependence, no surprise bills.
Security-first
Two decades of security discipline baked into every guardrail.
Local-first
Local models handle the work, so most runs cost exactly $0.
Open-source
The pieces worth sharing go public — measured and documented.
20+
yrs security
100%
self-hosted
Arc B60
battlemage gpu
OSS
open by default
Work With Me

hire the operator.

01
AI Automation & Agents

Browser and desktop agents that do real work — within guardrails you control.

02
Self-hosted Infra

Local-first AI infrastructure, cost-controlled and off the cloud treadmill.

03
Security Review

Two decades of security discipline, applied to your AI and systems.

Creator
Brandon Goolsby (he/him)

SENIOR AI ENGINEER · CYBERSECURITY

Builder of autonomous AI systems with a 20+ year cybersecurity background. I run a full self-hosted fleet on budget Intel Arc hardware — then open-source the parts worth sharing. Always innovating.

Security+ISC2 CCCCNA-levelSonicWall SNSA
LinkedIn ↗ GitHub ↗