Somnuss

Margetrp Hub

Building practical AI workbenches and self-hosted tools.

OpenAI-compatible workflows | Browser-first products | Deployable utilities

About

I build practical tools that turn scattered AI workflows into focused, usable products.

My recent work is centered on:

• AI image workbenches with OpenAI-compatible routes
• browser-first tooling that keeps the workflow in one place
• self-hosted utilities that are easier to deploy, test, and maintain
• clear boundaries between public UI, private infrastructure, and production data

Featured Projects

AI Image Workbench

A browser-first image creation workstation for OpenAI-compatible APIs and gateways.

It brings prompts, reference images, generation settings, canvas iteration, queues, and history into one focused workspace instead of spreading the workflow across multiple tools.

GPT Account Manager

A self-hosted operations workspace for mailbox intake, maintenance flows, queue handling, and account support tooling.

This project is more operations-oriented than the image workbench, so I treat deployment boundaries, data handling, and environment separation as first-class concerns.

What I care about

• products that solve real workflow friction
• interfaces that stay simple under heavy use
• deployable open-source tooling, not just demos
• careful security and data-boundary thinking

Current direction

Right now I am pushing toward a more complete stack around practical AI tooling:

• better workstation UX
• cleaner OpenAI-compatible integrations
• stronger self-hosted deployment paths
• more reliable long-running maintenance flows

Notes

• Some repositories are Chinese-first because that is where they were actively used first.
• Public repositories do not include private production keys or private backend systems.
• If you are building around similar workflows, issues and discussions are welcome.

Collapse