What is PLATO Torch?

PLATO Torch provides self-training rooms — thematic environments that accumulate knowledge (tiles) from agent interactions and learn from them automatically. Each room is a learning system that gets smarter with use.

from plato_torch import PRESET_MAP

# 26 training room presets available
room = PRESET_MAP["supervised"]()
room.feed(data={"features": [1, 2, 3], "label": "positive"})
room.train_step()
prediction = room.predict(input={"features": [4, 5, 6]})

Training Room Presets

Core Concepts

• Tile — Atomic knowledge unit (question/answer/domain/confidence)
• Room — Collection of tiles with a training strategy
• Room Sentiment — The room reads its own vibe and steers exploration
• Ensign — Compressed instinct distilled from a room (via plato-ensign)

Installation

pip install plato-torch

Architecture

Agent Interaction
       │
       ▼
   🧱 TILE (atomic knowledge)
       │
       ▼
   🏛️ ROOM (self-training collection)
       │
       ▼
   🎖️ ENSIGN (compressed instinct)
       │
       ▼
   [Any Model] — instant domain expertise

The Deadband Protocol ensures every room trains on safe channels:

• P0: Map negative space (where NOT to go)
• P1: Find safe channels
• P2: Optimize within bounds

For Agents

plato_torch_v1:
  type: training_room_library
  presets: 26
  input: tiles (Q/A/domain/confidence)
  output: trained_room → ensign
  deadband: P0→P1→P2
  install: "pip install plato-torch"

License

MIT