Technical University of Munich × SAP
AI Research / Multi-Agent Systems · 2026
Mycellium: Orchestrator-Free Multi-Agent Swarm System
Led product ownership and architecture for a novel orchestrator-free multi-agent system built in the TUM × SAP Practical Course.
Context
Starting point and environment
Many multi-agent systems rely on central orchestrators that can become bottlenecks and single points of failure. This project explores an emergent coordination approach where agents self-organize through shared state and local decision rules.
Challenge
Core constraint to solve
Design a robust orchestrator-free system where agents can coordinate, self-select tasks, and converge on reliable outputs without centralized control, while remaining understandable to technical and non-technical stakeholders.
Solution
Design and implementation approach
Designed the full system architecture and led a team of 6 through research, implementation, and stakeholder demos. The system combines capability- and trust-weighted bidding, an append-only blackboard, and consensus built through agents citing and building on each other's findings. Defined agent contract structures, tool interfaces, and interaction protocols, and built a visual canvas UI to configure networks and replay swarm runs step by step.
Outcome
Measurable impact and delivery result
Delivered a working orchestrator-free multi-agent platform with interactive simulation tooling and presented the approach to SAP industry stakeholders as part of the TUM practical course.
System architecture
How the swarm reaches consensus.
Hover or focus a node for details.
The Plan
There's no central manager. A lightweight Market Maker reads the request and turns it into a plan, but it never decides who does the work or what's correct.
The Auction
Specialist agents are invited to bid on the plan. Each one judges its own fit and opts in or out. No one is assigned work against its will.
The Board
Winning agents post findings to a shared board and cite each other's work. An idea only becomes a confirmed finding once enough other agents independently back it.
The Answer
One agent drafts a final answer from the confirmed findings, but even that draft only stands if the rest of the swarm votes to approve it.
Orchestrated vs. orchestrator-free
| Failure mode | One bad decomposition breaks everything downstream | Harder to trace, but no single point of failure |
| Coordination | Central planner assigns work | Capability- and trust-weighted bidding + append-only blackboard |
| Consensus | Whatever the coordinator decides | Claims survive only if other agents independently cite them |
Key Takeaway
“Consensus doesn't need a judge, it needs a threshold: claims that earn independent citations survive, the rest quietly fade. Clear contracts and a visible simulator made both delivery and reviews much faster.”
Working on something similar? I take on select consulting and freelance work in agentic AI, RAG systems, and LLM reliability.