flywheel-engine

The compounding intelligence loop: every exchange becomes a tile, tiles inject context, the system compounds.

The Flywheel

1. Capture — Agent interactions generate raw knowledge
2. Refine — Raw knowledge becomes structured tiles (via tile-refiner)
3. Inject — Tiles inject into agent context windows via PLATO rooms
4. Compound — Agents build on previous knowledge, generating deeper tiles
5. Loop — Each cycle produces more valuable knowledge than the last

Installation

pip install flywheel-engine

Usage

from flywheel_engine import Flywheel, Tile, Room

room = Room(name='fleet_orchestration')
tile = Tile(question='How do agents coordinate?', answer='Via Bottle Protocol...')
room.add_tile(tile)

flywheel = Flywheel()
flywheel.add_room(room)
result = flywheel.compound()  # Returns compounded knowledge

Architecture

The flywheel is the core of the Cocapn fleet's self-improving knowledge system. It transforms agent interactions into persistent, queryable knowledge that compounds over time.

Part of the Cocapn Fleet

Works with tile-refiner (raw→structured), plato-tile-spec (format), and plato-server (storage).

License

MIT