# Multi-Agent Systems

> An interactive, bilingual (EN/DE) reference for the patterns, tradeoffs, and failure modes of LLM-based multi-agent architectures. Covers a four-level complexity ladder (single agent → workflow → multi-agent → production), a pattern catalog, framework comparisons, agent-communication protocols, anti-patterns, and production concerns.

## Reference
- [Patterns overview](https://multiagentsystems.sites.markushanses.de/en/pattern/overview/): the full pattern catalog, organized by domain.
- [Coordination Patterns](https://multiagentsystems.sites.markushanses.de/en/pattern/coordination/): the six foundational patterns.
- [Social patterns](https://multiagentsystems.sites.markushanses.de/en/pattern/social/): multi-agent collaboration topologies.
- [Operational patterns](https://multiagentsystems.sites.markushanses.de/en/pattern/operational/): production-time system operation (memory, tools, runtime, governance, evals).
- [Agentic design patterns](https://multiagentsystems.sites.markushanses.de/en/pattern/agentic/): the ADP groups.
- [Anti-patterns](https://multiagentsystems.sites.markushanses.de/en/pattern/anti/): common failure modes and their fixes.

## Frameworks & Protocols
- [Frameworks](https://multiagentsystems.sites.markushanses.de/en/frameworks/): LangGraph, CrewAI, AutoGen, and others compared by native support.
- [Protocols](https://multiagentsystems.sites.markushanses.de/en/protocols/): agent-communication protocols (MCP, A2A, and more).

## The Ladder
- [L1 — Single agent](https://multiagentsystems.sites.markushanses.de/en/ladder/l1/)
- [L2 — Workflow](https://multiagentsystems.sites.markushanses.de/en/ladder/l2/)
- [L3 — Multi-agent](https://multiagentsystems.sites.markushanses.de/en/ladder/l3/)
- [L4 — Production](https://multiagentsystems.sites.markushanses.de/en/ladder/l4/)

## Decide & Learn
- [Decision guide](https://multiagentsystems.sites.markushanses.de/en/perspective/decide/): how to choose a pattern.
- [Determinism](https://multiagentsystems.sites.markushanses.de/en/perspective/determinism/): why not one LLM.
- [Learning paths](https://multiagentsystems.sites.markushanses.de/en/perspective/learn/): guided ramps.

## Lookup
- [Glossary](https://multiagentsystems.sites.markushanses.de/en/glossary/): agentic-engineering terms.
- [Sources](https://multiagentsystems.sites.markushanses.de/en/sources/): the bibliography.
- [Production concerns](https://multiagentsystems.sites.markushanses.de/en/production/): memory, tools, runtime, governance, and evaluation concerns for running agents in production.

## Full content
- [llms-full.txt](https://multiagentsystems.sites.markushanses.de/llms-full.txt): the complete reference text (knowledge base + glossary) for ingestion.