The funding and who backed it

The €5.7 million pre-seed round (roughly $6 million) was co-led by Cherry Ventures and Redstone. Angel Invest, Linden Capital, and additional business angels also participated, making this INXM’s first disclosed outside funding. The company plans to use the capital to grow its team, accelerate product development on the Orchestrator engine, and expand customer acquisition among European industrial enterprises.

The founding team: from rockets to reliable workflows

INXM was founded in 2025 by four people who spent their previous careers making extremely complex, safety-critical systems work predictably:

• Alexander (Alex) Oelling, CEO — formerly Chief Digital Officer at Isar Aerospace and Volocopter.
• Matthias Kainer — built launch orchestration systems at Isar Aerospace.
• Jesper Bylund — designed flight control interfaces and previously worked at automation platform n8n.
• Kamil Klüber — a veteran of software consultancy Thoughtworks.

That aerospace DNA is not a biographical footnote—it explains the company’s entire technical philosophy. Launch software must execute deterministic sequences under strict validation; it cannot improvise. The team applied the same mental model to enterprise AI.

Why standard enterprise AI agents fail

Most LLM-based automation tools invoke the language model at every step of a workflow. That creates three hard problems for regulated businesses:

• Non-determinism. The same prompt can produce different actions or outputs on different runs, making compliance and audit trails nearly impossible.
• Hallucination risk. A model that invents an approval step or skips a safety check creates unacceptable liability in manufacturing or finance.
• Data sovereignty. Sending sensitive production data to third-party LLM APIs conflicts with GDPR, the EU AI Act, and internal security policies.

The result is that many industrial companies evaluate AI agents but never deploy them in production. The technology looks powerful in demos but cannot meet the reliability bar that factory-floor software has enforced for decades.

How Compiled AI works differently

INXM’s Orchestrator engine borrows the architecture of a traditional software compiler rather than a chatbot loop. The process breaks into two distinct phases:

1. Compilation phase. A user describes a business process in natural language. An LLM generates a complete, structured, executable “Plan”—essentially deterministic code—and that plan is validated before it is ever allowed to run. The model is used once, during authoring, not at transaction time.

2. Execution phase. The validated Plan runs on INXM’s Orchestrator engine without calling an LLM again. The engine coordinates across the enterprise’s existing systems—ERP, PLM, MES, email, and approval tools—executing the same steps in the same order every time.

The technical difference is stark: standard agents invoke the LLM on every step (non-deterministic, token-expensive, hard to audit). INXM compiles once, then runs a fixed program. Outputs become reproducible, testable, and free from runtime hallucinations. The company calls this the flexibility of natural language combined with the reliability of traditional code.

To make the concept concrete, consider a quality-inspection workflow in a factory: an engineer describes the required approval steps and data checks in plain language. Orchestrator compiles that into a fixed sequence, a manager approves it once, and the workflow then runs identically every time a relevant event fires—without ever sending data outside the building or risking a model improvisation.

Infrastructure, compliance, and industrial targets

Because Orchestrator runs on the enterprise’s own infrastructure—on-premises or in a private cloud—production data never leaves the building. That architecture keeps the system compliant with GDPR and the EU AI Act by design. It also allows INXM to reach factory-floor software that cloud-only automation tools physically cannot touch.

The company is explicitly targeting industrial manufacturers and Germany’s Mittelstand: companies with legacy on-prem systems, regulatory obligations, and hard requirements for repeatability. This is not a horizontal automation play. It is purpose-built for enterprises that already rejected generic AI agents because they cannot be trusted in production.

With €5.7 million in fresh capital, a founding team that has shipped safety-critical software, and a technical approach that re-frames enterprise AI as a compiler problem rather than a conversational one, INXM is betting that the industrial world’s AI adoption bottleneck is not capability—it’s reliability.

The funding and who backed it

The €5.7 million pre-seed round (roughly $6 million) was co-led by Cherry Ventures and Redstone. Angel Invest, Linden Capital, and additional business angels also participated, making this INXM’s first disclosed outside funding. The company plans to use the capital to grow its team, accelerate product development on the Orchestrator engine, and expand customer acquisition among European industrial enterprises.

The founding team: from rockets to reliable workflows

INXM was founded in 2025 by four people who spent their previous careers making extremely complex, safety-critical systems work predictably:

• Alexander (Alex) Oelling, CEO — formerly Chief Digital Officer at Isar Aerospace and Volocopter.
• Matthias Kainer — built launch orchestration systems at Isar Aerospace.
• Jesper Bylund — designed flight control interfaces and previously worked at automation platform n8n.
• Kamil Klüber — a veteran of software consultancy Thoughtworks.

That aerospace DNA is not a biographical footnote—it explains the company’s entire technical philosophy. Launch software must execute deterministic sequences under strict validation; it cannot improvise. The team applied the same mental model to enterprise AI.

Why standard enterprise AI agents fail

Most LLM-based automation tools invoke the language model at every step of a workflow. That creates three hard problems for regulated businesses:

• Non-determinism. The same prompt can produce different actions or outputs on different runs, making compliance and audit trails nearly impossible.
• Hallucination risk. A model that invents an approval step or skips a safety check creates unacceptable liability in manufacturing or finance.
• Data sovereignty. Sending sensitive production data to third-party LLM APIs conflicts with GDPR, the EU AI Act, and internal security policies.

The result is that many industrial companies evaluate AI agents but never deploy them in production. The technology looks powerful in demos but cannot meet the reliability bar that factory-floor software has enforced for decades.

How Compiled AI works differently

INXM’s Orchestrator engine borrows the architecture of a traditional software compiler rather than a chatbot loop. The process breaks into two distinct phases:

1. Compilation phase. A user describes a business process in natural language. An LLM generates a complete, structured, executable “Plan”—essentially deterministic code—and that plan is validated before it is ever allowed to run. The model is used once, during authoring, not at transaction time.

2. Execution phase. The validated Plan runs on INXM’s Orchestrator engine without calling an LLM again. The engine coordinates across the enterprise’s existing systems—ERP, PLM, MES, email, and approval tools—executing the same steps in the same order every time.

The technical difference is stark: standard agents invoke the LLM on every step (non-deterministic, token-expensive, hard to audit). INXM compiles once, then runs a fixed program. Outputs become reproducible, testable, and free from runtime hallucinations. The company calls this the flexibility of natural language combined with the reliability of traditional code.

To make the concept concrete, consider a quality-inspection workflow in a factory: an engineer describes the required approval steps and data checks in plain language. Orchestrator compiles that into a fixed sequence, a manager approves it once, and the workflow then runs identically every time a relevant event fires—without ever sending data outside the building or risking a model improvisation.

Infrastructure, compliance, and industrial targets

Because Orchestrator runs on the enterprise’s own infrastructure—on-premises or in a private cloud—production data never leaves the building. That architecture keeps the system compliant with GDPR and the EU AI Act by design. It also allows INXM to reach factory-floor software that cloud-only automation tools physically cannot touch.

The company is explicitly targeting industrial manufacturers and Germany’s Mittelstand: companies with legacy on-prem systems, regulatory obligations, and hard requirements for repeatability. This is not a horizontal automation play. It is purpose-built for enterprises that already rejected generic AI agents because they cannot be trusted in production.

With €5.7 million in fresh capital, a founding team that has shipped safety-critical software, and a technical approach that re-frames enterprise AI as a compiler problem rather than a conversational one, INXM is betting that the industrial world’s AI adoption bottleneck is not capability—it’s reliability.