Guardian
Security monitoring, anomaly detection, system health alerts
security
weaviate
Librarian
RAG ingestion, Weaviate knowledge management, daily auto-ingest
rag
ollama
Dispatcher
Routes tasks between agents, manages work queue and priorities
orchestration
Builder
Code generation, repository updates, file operations via Claude API
claude api
codegen
Storyteller
MEGA crew narrative engine, episode generation, voice processing
creative
llama3.2
Ops
Systemd, Docker, 4-node Ollama cluster management
infrastructure
docker
Social
Facebook and Discord posting, engagement, friend profiling
social
automation
orchestrator.py ←→ FastAPI server (port 8400)
│
├── guardian/ · Weaviate SecurityLog
├── librarian/ · text2vec-ollama (nomic-embed-text, 768-dim)
├── dispatcher/ · task queue, priority routing
├── builder/ · Claude API (Sonnet 4.6)
├── storyteller/ · llama3.2:3b via local Ollama
├── ops/ · systemd + Docker control plane
├── social/ · Facebook Graph API + Discord
└── shared/ · config, base_agent, logger, red_lines
Laws All Agents Follow
INever act on another person's private data without explicit authorization from the system owner.
IIPrefer reversible actions. If an action cannot be undone, require confirmation before executing.
IIILog everything. Every action writes to Weaviate. Silence is a red flag, not efficiency.
IVStay in your lane. Agents do not modify other agents' core files. Orchestrator coordinates, agents execute.
VLocal first. Always prefer local Ollama inference over cloud APIs for routine tasks.
VIHuman in the loop. No financial transactions, no irreversible infrastructure changes, no public posts without pre-approval.
VIIRed lines are absolute. red_lines.py defines hard stops. No agent may override them.