The cosmological hierarchy

Galaxy → Sun → Planet → Biome → Stardust → Entity → Nebula. Seven layers, each with distinct lifecycle and query semantics.

Orion organizes knowledge into seven nested layers. The metaphor is cosmological, but the mapping to real structures is precise:

Galaxy                          (your entire knowledge universe)
├── Sun                         (steering document — 7 sections)
├── Planets                     (knowledge domains)
│   └── Biomes                  (project contexts)
│       ├── Stardust            (atomic knowledge records)
│       └── Entities            (named concepts, linked into graph)
└── Nebula                      (activity log)

Layer	Maps to	Lifecycle	Example
Galaxy	Organization or individual	Permanent	"Acme Corp" or "Andy's Galaxy"
Sun	Steering document	Evolves via evolution_log	Identity, values, agent protocol
Planet	Department or domain	Permanent	Engineering, Product, Personal
Biome	Project or context	SEED → ACTIVE → MATURE → DORMANT → ARCHIVED	"Auth Refactor", "Q3 Migration"
Stardust	Atomic knowledge record	Confidence decays without access	A decision, fact, or procedure
Entity	Named concept	Tier promotes with mentions (1→2→3)	PostgreSQL, FastAPI, "Sarah"
Nebula	Activity log	Append-only	Every read, write, and audit event

Why the hierarchy matters

The hierarchy determines scope through gravity levels. When you store knowledge, you choose how broadly it applies:

• BIOME gravity — relevant only within this project. "The auth refactor uses session tokens."
• PLANET gravity — relevant across the entire domain. "We use PostgreSQL for all services."
• GALAXY gravity — universal. "Always use UTC timestamps."

Gravity affects retrieval: memory.context builds context bundles by pulling GALAXY-gravity records first, then PLANET, then BIOME — ensuring universal rules are always present regardless of token budget.

Key design decisions

One Sun per Galaxy. The Sun is the single source of truth for agent behavior. It has 7 sections (identity, values, agent_protocol, planet_registry, working_context, steering_doc, evolution_log) and every change is logged. Planets can override agent_protocol for domain-specific rules, but the Sun is always the default.

Biomes have lifecycles. A Biome starts as SEED when created, advances to ACTIVE on first write, MATURE after sustained activity, DORMANT after 30 days of inactivity, and ARCHIVED when explicitly closed. The weekly audit manages transitions automatically.

Every Galaxy has an Inbox. When the routing engine can't confidently assign knowledge to a Planet, it goes to the Inbox Biome for manual review. This prevents silent misrouting — you always know when the system is uncertain.

Entities are auto-extracted. You never manually create entities. They emerge from stardust content via regex patterns and optional LLM extraction, then link into the knowledge graph automatically.