I'm Sterling Schwieger — a software engineer who designs and builds end-to-end systems. My flagship work is Glass: a self-hosted, AI-native MSP platform of 16 microservices, local LLM inference, and an integrated security engine — architected and written solo.
Glass is a unified platform for managed service providers — ticketing, RMM, CMDB, billing, security, telephony, and AI triage — that replaces a stack of commercial SaaS tools like ConnectWise, Kaseya, and NinjaOne. Every byte stays on your own infrastructure, and every AI inference runs on a local GPU.
Ticket triage, email classification, voicemail transcription, and NLP→SQL reporting run on a local DGX Spark via Ollama. No client data ever leaves the network.
A sandbox model screens every untrusted input for prompt injection before the privileged model sees it — an isolation pattern no commercial MSP platform ships.
11 specialized dashboards — from an RMM matrix to a 3D facility viewer — unify tickets, assets, alerts, billing, and network topology in one UI.
16 independently deployable services in PHP, Go, and Python, orchestrated with Docker Compose profiles and JWT/OIDC auth delegated to a central identity service.
768-dimension embeddings in pgvector power cosine-distance search across tickets and knowledge — surfacing the top similar cases at the moment a ticket is opened.
SSO/OIDC, mandatory admin MFA, a Go-based XDR engine (Wazuh + Rspamd + ClamAV), CRU CA mutual TLS, and a full Loki/Grafana observability pipeline.
Glass reaches an estimated ~94% feature parity with the major MSP platforms while keeping data sovereignty, source-code auditability, and freedom from supply-chain risk that centralized SaaS can't offer.
# Glass — profile-based orchestration services: glass-app: # PHP 8.3 monolith · 11 dashboards depends_on: [glass-core, glass-db, redis] glass-ai-worker: # Python · Dual-LLM sandbox + RAG image: ollama/local-inference glass-sentinel: # Go · multi-tenant XDR engine ports: ["8090"] glass-tickets: # Slim 4 · SLA sidecar · pub/sub glass-crm: # Go 1.25 + React · Redis Streams # …11 more services, 5 sidecars, 8 infra # $ docker compose --profile all up -d # ✔ 29 containers · healthy
From database schema and Go concurrency to React frontends, LLM pipelines, and production DevOps — I design, build, and operate the entire system.
PHP 8.3 (Slim 4), Go 1.25, Python/FastAPI. REST + GraphQL, JWT/OIDC auth, event-driven pub/sub, and queue-based async processing.
PostgreSQL 16, pgvector embeddings, RAG pipelines, local LLM inference with schema-enforced outputs, and Redis caching/streams.
React 18 + TypeScript, Vite builds, WebSocket/WebRTC, Docker Compose, Nginx + mTLS, Cloudflare Tunnel, and CI-style automated deployments.
Glass is my flagship. It lives alongside a family of companion CRU products — built by my colleague Blake Thoeness — that together form a full self-hosted platform for security, backup, and AI infrastructure.
A 16-microservice, AI-native operating system for a managed service provider — ticketing, RMM, CMDB, billing, security, and local LLM intelligence in one self-hosted stack.
A central authentication platform: an internal certificate authority with strict mTLS, a secrets/2FA vault (OpenBao "KeyMaster"), and an OIDC/SAML identity provider — one pane of glass for certs, keys, and SSO.
A cross-platform endpoint backup product — cloud console, agent fleet, and a Wails desktop app — with Carbonite-style Explorer status overlays, a Kopia engine fork, and mTLS data-plane ingress.
Command & control for AI agents across a heterogeneous GPU fleet — routing inference, code, voice, and transcription tasks to the best-fit node by live hardware telemetry, with fleet self-update.
A high-performance gateway that unifies OpenAI, Anthropic, Gemini, and OpenAI-compatible providers under one endpoint — protocol translation, load balancing, failover, and usage tracking, transparently.