PLATO Demo — 5-Atom Reasoning Chain

🔮 cocapn-plato v0.2.0 — Knowledge system that reasons like a captain.

This demo shows PLATO's 5-atom reasoning chain applied to a practical question: "What is the optimal team size for a fishing crew?"

What It Demonstrates

The 5 atoms (each a real LLM call, NOT a template):

Atom	Role	Description
1	PERCEIVE	Read the room, identify what's known vs. gaps
2	ANALYZE	Break into constraints, variables, stakeholders
3	REASON	Derive the answer from the analysis
4	REFINE	Critical review — edge cases, assumptions
5	SUBMIT	Synthesize final knowledge tile

Each atom calls an actual LLM. The chain shows tokens consumed per atom, proving real inference.

Quick Start

# Prerequisites: set at least one API key
export DEEPSEEK_API_KEY="sk-..."   # recommended
# or
export GROQ_API_KEY="gsk_..."

# Run the demo
python3 demo.py

# Or specify provider
python3 demo.py "What is the best tide for fishing?" deepseek
python3 demo.py "What is the best tide for fishing?" groq

Output

The demo prints the full reasoning chain to stdout and saves to output.txt:

======================================================================
  PLATO 5-ATOM REASONING CHAIN
======================================================================
  Question: What is the optimal team size for a fishing crew?
  Provider: deepseek (deepseek-chat)
======================================================================

── ATOM 1: PERCEIVE ──────────────────────────────────────────
  LLM call: YES | tokens: 431

**KNOWN**
...

── ATOM 2: ANALYZE ───────────────────────────────────────────
...

── ATOM 3: REASON ─────────────────────────────────────────────
...

── ATOM 4: REFINE ─────────────────────────────────────────────
...

── ATOM 5: SUBMIT ─────────────────────────────────────────────
...

  Chain Stats:
    Atoms: 5/5
    LLM calls: 5 ← real inference, NOT templates
    Total tokens: 8,155
    Provider: deepseek/deepseek-chat

Record a Demo GIF

Option A: asciinema (recommended)

# Install
pip install asciinema

# Record (from project dir)
asciinema rec /tmp/plato-demo recording.json
# ... run demo: python3 demo.py ...
# Exit with Ctrl+D or Ctrl+C

# Convert to GIF
pip install agg
agg /tmp/plato-demo/recording.json /tmp/plato-demo/demo.gif

Option B: byexample + termtosvg

pip install byexample termtosvg
byexample run -l python3 demo.py --asciinema /tmp/plato-demo/recording.json
svg-term --input /tmp/plato-demo/recording.json --output /tmp/plato-demo/demo.gif --window

Option C: Quick terminal capture

# Scripted — no interactivity needed
script -q -c "python3 demo.py" /tmp/plato-demo/terminal.out
# Then convert .out to GIF with agg or share as-is

Project Structure

plato-demo/
  demo.py          ← Main demo (5-atom chain)
  output.txt       ← Captured output from run
  README.md        ← This file
  record.sh        ← Recording script

Architecture

demo.py
  └── TrainedReasoner          (wraps a simple agent)
        └── APIClient          (DeepSeek/Groq/SiliconFlow)
              └── 5× LLM calls (PERCEIVE → ANALYZE → REASON → REFINE → SUBMIT)

Provider Support

Provider	Env Variable	Best Model
DeepSeek	DEEPSEEK_API_KEY	deepseek-chat
Groq	GROQ_API_KEY	llama-3.3-70b-versatile
SiliconFlow	SILICONFLOW_API_KEY	deepseek-ai/DeepSeek-V3

Related

• plato-sdk — Full Python SDK with armor, skills, equipment layers
  https://github.com/SuperInstance/plato-sdk
• plato-server — Standalone PLATO server with HTTP API
  https://github.com/SuperInstance/plato-server
• plato-mud-server — Multi-user dungeon PLATO backend
  https://github.com/SuperInstance/plato-mud-server

---

Part of the Cocapn fleet — agents that live in PLATO, learn from each other, improve for everyone.