Productizer
live
Requirements that stay true.
One living spec per product. Work stops when something contradicts what you agreed.
- One living spec
- EARS requirements
- Claude Code plugin
AI agents are becoming theoperator
MoorAI
live
Guardrails that run on the machine.
Every prompt is reviewed locally, before it reaches the agent — coach, alert or block by policy.
- On-device
- Coach · alert · block
- Content-free signals
AppCrane
live
A home for the apps your AI builds.
The agent is the operator. The platform keeps the receipts — human or agent, every action has a name on it.
- Self-hosted
- Docker isolation
- SSO + per-user audit
IT-AI
live
Remote IT support you host yourself.
Diagnoses your fleet and, with your approval, repairs it.
- Read-only by default
- Screen · shell · files
- Nothing leaves your net
Four tools, one order of operations. Hover any of them.
What every tool shares
Self-hosted
Runs on your own server. No relay, no middleman, no metered egress.
Agent-native
Driven over the open Model Context Protocol — from the agent you already use.
Open source
Read every line, fork it, run it — nothing hidden in a cloud. MoorAI is MIT; AppCrane is AGPL-3.0.
Private by design
Data and prompts stay on your machines. Nothing phones home.
// the portfolio
Four tools, one principle
Build
Productizer
Track requirements as a living spec. Every request is checked against everything already agreed — and when one contradicts another, the work waits for your ruling instead of quietly overwriting it.
Guard
MoorAI
AI security at the point. Open-source runtime guardrails for AI agents, apps and APIs — coding agents such as Claude Code, Codex, Copilot CLI, Gemini CLI and Cursor, Agent SDK services, any framework through a sidecar, MCP servers and model APIs. File reads, commands, tool calls and model calls are checked before they run: coach, alert, or block by policy. Security teams get redacted, content-free signals, never the prompts — governance without surveillance.
- local-first — Every prompt reviewed on-device — by default, never leaves the machine.
- policy — Coach, alert, or block by policy, before the agent runs.
- content-free — Security teams get redacted signals, never the prompt text.
- context — Catches a secret or PII being read into the agent's context — a .env slurped by the agent, not just typed in a prompt.
- tool calls — The hook denies the tool call even when the prompt itself was never classified.
- open source — The agent is MIT-licensed; the management server is source-available under the Elastic License 2.0.
Deploy
AppCrane
A home for your AI apps. Deploy and run the apps your AI builds on a server you own — where your code, data, and per-user audit trail never leave your infrastructure. GitHub-native flow, Docker isolation, SSO — the one thing managed clouds can't offer.
Control
IT-AI
Self-hosted, AI-powered remote IT support for your own fleet. An AI that sees a machine, diagnoses the problem, and — only after a human approves — fixes it. Screen, shell, files and live device health from one hub you run. Read-only by default; nothing changes until someone clicks Apply, and nothing leaves your infrastructure.
- diagnose-fix — An AI that diagnoses the problem — and fixes it, once you approve.
- approval-gated — Read-only by default; every change waits for a human to click Apply.
- self-hosted — Your hub, your machines — screen, shell and files never leave your infrastructure.