plato-ensign 0.1.0

PLATO Ensign — room-trained instincts for fleet agents

🎖️ PLATO Ensign

Compressed instincts from PLATO rooms. Load onto any model.

Part of Cocapn — Agent Infrastructure for Intelligence.

---

What is an Ensign?

An ensign is a compressed instinct distilled from a PLATO room's accumulated knowledge. Think of it as a domain expertise adapter that loads onto any model — not a full LoRA, but a portable knowledge packet that gives instant competence.

from plato_ensign import EnsignLoader

# Load the fleet health ensign onto any model
loader = EnsignLoader()
ensign = loader.load("fleet-health")

# The ensign contains distilled knowledge from hundreds of tiles
# across the fleet-health room
print(ensign.summary)  # "Fleet monitoring patterns, service health checks..."
print(ensign.quality)   # 0.87
print(ensign.source_tiles)  # 142 tiles distilled

How It Works

PLATO Room (accumulates tiles)
       │
       ▼
  Room Trainer (synthesizes room knowledge)
       │
       ▼
  Ensign Export (compressed instinct)
       │
       ├── JSON (interpretable, any model)
       ├── LoRA adapter (fine-tuned weights)
       └── GGUF metadata (edge deployment)

Ensign Types

Type	Size	Load Time	Best For
JSON	~10KB	Instant	Context injection, prompt augmentation
LoRA	~100MB	~2s hot swap	Behavioral adaptation
GGUF	~50MB	~1s	Edge deployment (Jetson)

The Flywheel

Agents work → produce tiles → rooms accumulate
       → room trainer distills → ensigns exported
       → ensigns loaded back into agents → agents work better
       → better tiles → better rooms → better ensigns
       → COMPOUNDS EVERY CYCLE

For Agents

plato_ensign_v1:
  type: compressed_instinct
  input: room_tiles (accumulated)
  output: ensign (JSON/LoRA/GGUF)
  quality_range: [0.0, 1.0]
  load_time: "<2s (LoRA hot swap)"
  universality: "any model, any device"

License

MIT