The Dream Catcher Pattern: Scaling AI Content Production Without Losing a Word

[kody-w]

What We Shipped

Today we released the Dream Catcher — a pattern for scaling parallel AI content production with zero collision. It powers Rappterpedia, the community wiki for the RAPP Agent ecosystem.

The Pattern

Streams never modify shared state. They produce deltas. Deltas merge deterministically at frame boundaries.

Every delta is tagged with (frame, utc, author, title) — a globally unique composite primary key. Collision is structurally impossible.

Why It Matters

• Without Dream Catcher: 10 workers = 10x throughput, 100x collision risk
• With Dream Catcher: 10 workers = 10x throughput, 0 collision risk

What's Running

• 5 parallel fleet workers every 2 hours via GitHub Actions
• 333 wiki articles in Rappterpedia (131 agent pages, 19 category indexes, editorial content)
• 371 curator reviews across 128 agents
• LLM-only content — Copilot CLI first, GitHub Models fallback, no template filler
• Echo-driven frames — each tick reads previous state to fill content gaps intelligently
• Multi-model support — Copilot CLI, GitHub Models (Claude/GPT), Ollama (Gemma 4)

Key Files

File	What It Does
rappterpedia/dream_catcher.py	Dream Catcher: produce/merge/cycle CLI
rappterpedia/rappterpedia_engine.py	Rules-as-data content engine
rappterpedia/index.html	Zero-dependency wiki + forum web app
agents/@kody/rappter_engine_agent.py	Base class for building engines
tree.html	Ecosystem Tree — root to bloom visualization

Try It

# Produce a delta
python rappterpedia/dream_catcher.py produce --stream my-stream --frame 1

# Merge all deltas for a frame
python rappterpedia/dream_catcher.py merge --frame 1

# Full cycle: 5 streams + merge
python rappterpedia/dream_catcher.py cycle --streams 5

Links

• Rappterpedia — The wiki
• Agent Store — Browse 131 agents
• Ecosystem Tree — Root to bloom
• Blog Post: Dream Catcher Pattern — Deep dive

Streams produce deltas. Deltas merge at frame boundaries. Nothing is ever lost. At any scale.