This repository contains runtime data for the Mind-Swarm M-brain AI agent orchestration platform.

Runtime contains M-brain agent definitions and modules that can be dynamically loaded while the core server is running:

• /mbrain_agent_types - M-brain agent type definitions (YAML)
• /dspy_modules - M-brain DSPy modules and tools
• /knowledge - Knowledge packs for M-brain agents

Mind-Swarm uses a clean separation between:

1. mind-swarm-core - The stable execution engine (server, infrastructure, storage, auth, etc.)
2. mind-swarm-runtime - This repo - M-brain definitions and modules

Mind-Swarm now exclusively uses the M-brain (Multi-brain) architecture:

• Unified Memory Model: Single memory system shared across all brain channels
• Brain Channels: Specialized processing pipelines (Knowledge, Doer, Fixer, etc.)
• DSPy-based: All brains are implemented as DSPy modules
• No Templates: M-brains generate responses dynamically without Jinja2 templates
• No Legacy Tools: M-brains use DSPy tool modules in /dspy_modules/mbrain_modules/tools/

Current M-brain types in /mbrain_agent_types/:

• resilient - General purpose resilient agent
• root - System root agent
• ui - User interface agent
• user_assistant - User's personal assistant

Knowledge packs in /knowledge/ provide:

• Agent-specific knowledge
• Brain-specific knowledge
• Tool documentation
• System overview and glossary

• ✅ Pure M-brain architecture
• ✅ No legacy agents or project infrastructure
• ✅ DSPy-based tool system
• ✅ Knowledge pack integration