Launching INXM: Alexander Oelling on Building Sovereign AI for Europe's Industrial Backbone
My Journey to INXM: From Rockets to AI Revolution
First off, a quick rewind on my background because it ties directly into why I started INXM. I've been in the tech and aerospace game for years—co-founded a couple of startups early on, then jumped into Volocopter where we built massive software systems with a huge team. This was pre-AI boom, so think old-school coding marathons. After that, I spent time as Chief Digital Officer (CDO) at Isar Aerospace SE, where we launched that stunning rocket from a fjord in Norway. If you saw the live stream, you know what I mean—it was like something out of a sci-fi movie, with all of Europe watching.
At Isar, my role as CDO was all about digital transformation in a super complex environment. Rockets aren't just assembled; they're precision-engineered beasts with tens of thousands of parts, most made in-house. We're talking about digitizing everything from production workflows (no more guys running around with paper checklists) to procurement, design approvals, and even ground control software. For example, during testing, we'd generate terabytes of data—analyzing that required custom software to handle simulations, risk assessments, and real-time monitoring. I built teams to create in-house tools that kept everything traceable and scalable, without relying on external vendors. That's a huge edge for Germany in space tech.
But here's where the spark for INXM came from. At Isar Aerospace, we were pushing boundaries, but we still dealt with industry standard systems and the need for quick, pragmatic solutions. Time was never on our side to fully roll out massive platforms. Instead, we leaned on cutting-edge tech to bridge gaps, and that's when AI—specifically agentic AI—started clicking for me. The FOMO (fear of missing out) in the AI world is real; there's so much innovation, investor buzz, and customer demand that I knew this was the moment to found something new. Enter INXM: a startup still in stealth mode, but we're building fast.
What INXM Is All About:
Sovereign, On-Prem AI Agents for the Real World
INXM stands for Intellectus Ex Machina—okay, that's my shorthand, but really, it's about giving European companies back their data sovereignty in the AI era. Think of it as next-gen AI agents that run locally, on your own infrastructure, complying with EU regs like the AI Act and GDPR, and will only do exactly what they are allowed to do. No shipping sensitive data to US servers; no black-box mysteries where you wonder if your files are getting tweaked or deleted without oversight.
The problem we're solving? Most AI tools today are cloud-based, often from Big Tech, and they compromise control. As CDO, I'd ask: Why can't I buy software that runs in my data center, integrates seamlessly with my processes, and delivers repeatable results? For instance, can’t risk sending rocket engine designs to ChatGPT—employees might do it accidentally if not guided, but that's a sovereignty nightmare. Important companies, especially in Germany's Mittelstand (mid-sized firms), still run half their standard software instances on-prem because they value control. Yet, they're missing out on AI because current options feel risky or unreliable.
Our solution: AI agents that support daily work without going rogue. These aren't wild, uncontrolled systems; they're guardrailed, traceable, and user-focused. Imagine an agent with access to your enterprise data (emails, SharePoint, OneDrive) but only acts on your commands. For example, a manufacturing firm gets a 200-page RFP (request for proposal) from an customer. Normally, a team scrambles for two weeks, risking oversights that cost millions. With INXM, you prompt: "Analyze this doc, run risk assessments, and draft our bid." It pulls from your supply chain data, formats everything consistently, and previews in a virtual sandbox—done in minutes, not days. Another example: "Prep my meetings today with email summaries and attached docs." It scans your inbox, compiles insights, but never alters originals without approval. And more importantly - when the way the ai worked led to good results you can store this as a plan, and rerun it again to always get the same great results, in the same format, in the same way.
We're building this with a top-tier team and using open-source components for transparency. Weekly sprints, partnerships with devs and consultants, and we're eyeing a downloadable software soon. No more hiring expensive low-code consultants (think 60k-150k euros per workflow); this is vibe-coding for enterprises—natural language interactions with your SAP, like "Book this incoming invoice and prep for approval," slashing 80 clicks to a simple chat.
The Agentic AI Hype vs. Reality: Why We're Focusing on Reliability
Let's get real about agentic AI—systems that plan and execute complex tasks autonomously. There's tons of buzz from players like OpenAI, Anthropic, and Perplexity, but the reality? Most projects flop because results aren't reliable. You get mixed signals: hype about AGI (artificial general intelligence) tomorrow, but tools that take 20 minutes for a simple email summary or expose your IP.
Our take: Reliability comes from structure. We use multi-stage mechanisms—traceability (see exactly where data comes from), guardrails (only allow safe actions), sandboxing (test in isolated environments), and static connectors (hand-coded by pros for consistent data exchange). Combine that with memory graphs and executable "recipes" (pre-planned steps), and you avoid the "soup" of dynamic chaos.
From our prototypes, this works wonders. For example, instead of letting AI freestyle a connector to your database (which might garble formats), we pre-define it. If the output doesn't match, it loops back until it does—ensuring repeatable, mission-critical results like API calls or bid preps. This isn't new science; agent systems were theorized in the early 2000s during my studies. But now, with advanced LLMs, we can finally build them right. Forward-thinking? Expect explosions of innovation soon—more interfaces, better sandboxes, virtual desktops—all on-prem. OpenAI does previews, but on their US servers; we're bringing it local.
Who Should Care? Ideal Customers and Massive Market Potential
From an investor lens, our ICP (Ideal Customer Profile) is clear: German Mittelstand firms with overflowing order books, thin margins, and talent shortages. These are producers handling high-volume contracts but struggling with repetitive tasks like documentation or enterprise bookings—no rocket science, but they need certified pros who are hard to find outside big cities.
Empower your existing team to work at a higher level. Automate the mundane for higher throughput without massive digital overhauls. Instead of millions on ERP expansions, dip into agentic AI with a simple IT admin. Market size? Huge—Europe's industry is ripe for this, especially with sovereignty pushes. We're not at the end; billions are pouring in, and systems like ours will democratize AI without consultant armies.
Wrapping this up, chatting about INXM felt like unleashing a bottled-up vision. Thanks to A11 for the platform—super exciting times ahead.
Podcast: https://www.podcast.de/episode/693045206/ki-agenten-in-der-industrie-mit-alexander-oelling