A construction group with 3 brands.
One platform. Everything visible.
a general contractor, a natural stone distributor, a window supplier — three companies under one roof. This is what we built for them, and it's running right now. Click through the live instances below.
Three brands. One operational challenge.
A German construction group managing three distinct brands — a general contractor (a general contractor), a natural stone distributor (a natural stone distributor), and a window supplier (a window supplier). Each brand has its own clients, its own products, its own website. But the back office is one team.
Their problem: everything ran on email, spreadsheets, and memory. Customer requests fell through the cracks. Product catalogs were PDFs. Nobody knew what the other brand was doing.
Websites that comply. Google Workspace that runs itself.
Each brand got a responsive website with full German compliance — Impressum, DSGVO privacy policy, cookie consent. Google Workspace was set up with managed users, groups, shared drives, MFA enforcement. Professional email for every team member.
Three brands. Three domains. One shared Google Workspace with organized drive structures and permission groups per brand.
Let clients tell you what they want.
Instead of guessing what features each brand's clients need, we built a Feature Voting Board. Clients submit feature requests in their own words. They vote on what matters most. AI enhances each submission into a proper user story with acceptance criteria, effort estimate, and cost projection. One click creates a GitHub issue in the right repo.
Each brand gets its own subdomain. Clients see only their brand's board.
This is the actual board running on our infrastructure. a general contractor clients submit and vote on features here. Try it — submit a feature request.
A memory system that actually knows your business.
We built a memory pipeline that ingests every conversation, extracts entities (people, companies, projects, products), maps relationships between them, and makes it all searchable by meaning — not just keywords.
The result: an AI assistant that remembers every client interaction, knows which project belongs to which brand, and can answer questions like "what did the a natural stone distributor client ask about last week?"
Live counts from our memory-api. Entities extracted via deterministic pattern matching (no LLM calls). Graph stored in PostgreSQL with pgvector.
5 agent runtimes. One dashboard.
Hermes is our primary agent — running as a systemd service with 5 MCP servers connected (memory-api, PostgreSQL, GitHub, Brave Search, filesystem). It has a 128K context window, compressed conversation history, and a session watcher that feeds new conversations into the memory pipeline every 2 minutes.
The dashboard shows agent health, memory ingestion rate, connected tools, and active sessions. Here's what's running right now.
AI that talks. Literally.
We deployed Dograh — an open-source voice AI platform — self-hosted on our own infrastructure. It gives each brand its own voice agent: phone-based scheduling, lead qualification, customer support. Connected to the CRM pipeline, so every call automatically creates or updates a contact.
Drag-and-drop workflow builder. No vendor lock-in. Runs on our server, not someone elses cloud. Which means the clients call data never leaves their infrastructure.
The actual Dograh instance running on our NetCup server. This is the same platform each client brand gets — configured with their workflows, their voice, their phone number.
What powers this
Everything you see here runs on two VPS servers — a Hostinger machine for the agent infrastructure and a NetCup machine for the client-facing services. No AWS. No enterprise subscriptions. Ollama Cloud for AI compute on a fixed monthly subscription instead of pay-per-token.
The same stack spins up for every new client. When you sign up, we don't start from scratch — we configure a proven system.
You don't start from zero.
Everything on this page — the feature voter, the memory pipeline, the agent dashboard, the CRM — is already built and running in production. When you sign up, we don't build from scratch. We configure this same stack for your brands, your clients, your data.
That's why setup is days, not months. And why we can show you a working demo on the first call.
Let's find the fix.
30 minutes. No pitch. We'll look at one workflow and tell you what to fix first.