最優先ルール（単一真実）：docs/厳格固定ルール.md 最新進捗：docs/status.md

Philosophy-Driven AI: When Pigs Fly

A frog in a well may not know the ocean, but it can know the sky.

Po_core is a philosophy-driven AI decision-support system. You give it a question; it returns structured options, reasons, counterarguments, uncertainty labels, and follow-up questions — all grounded in ethical deliberation.

What you get back:

• proposal — the winning response after multi-philosopher deliberation
• status — "ok" or "blocked" (3-layer safety gate result)
• options / questions — alternative framings and uncertainty probes
• trace — full audit log of which reasoning paths were taken and why

What Po_core is not:

• Not a truth oracle — it does not claim factual correctness
• Not an emotional-care chatbot — it provides structured reasoning, not emotional support
• Not a replacement for medical, legal, or financial judgment

pip install po-core-flyingpig

from po_core import run

result = run("What is justice?")
print(result["proposal"])   # Winning philosopher's response
print(result["status"])     # "ok" or "blocked"

Install note: Po_core pulls ML dependencies (torch, sentence-transformers, etc.). On a fresh environment the install may take several minutes. If torch or sbert are unavailable, the tensor backend falls back to tfidf/basic automatically — core deliberation still works.

Local development checkout:

git clone https://github.com/hiroshitanaka-creator/Po_core.git
cd Po_core
pip install -e ".[dev]"

Current AI optimizes for statistical accuracy — a brilliant parrot that understands nothing. Po_core asks a different question: what if AI reasoned from philosophy, not just data?

The differentiator: 42 philosophers (Western, Eastern, African, Canadian) run as interacting tensors through a 10-step hexagonal pipeline. Each brings its own reasoning module. They compete, interfere, and reconcile — producing a Pareto-optimal proposal with a measurable ethical pressure signal instead of a single confident prediction. (The internal dummy slot is a compliance sentinel helper and must not be counted as one of the 42.)

No matter how many relationships we have, decisions are made alone. That's why Po_core exists — to stand beside you when you must say "Leave it to me."

Read the full story: Manifesto

Hexagonal run_turn pipeline — 10 steps:

MemoryRead → TensorCompute → SolarWill → IntentionGate → PhilosopherSelect
→ PartyMachine → ParetoAggregate → ShadowPareto → ActionGate → MemoryWrite

Three tensor metrics measure the deliberation:

Safety is three-layered: IntentionGate (pre-deliberation) → PolicyPrecheck (mid-pipeline) → ActionGate (post-deliberation). SafetyMode transitions NORMAL → WARN → CRITICAL based on freedom_pressure thresholds.

For full source layout and component detail → Architecture docs · CLAUDE.md

Version 1.0.3 is published on PyPI. Package classifiers declare Development Status :: 4 - Beta.

For the complete evidence record, evidence gaps, and roadmap: docs/status.md

# From PyPI (recommended)
python3 -m venv venv
source venv/bin/activate  # Windows: venv\Scripts\activate
pip install po-core-flyingpig

# Specific version
pip install "po-core-flyingpig==1.0.3"

# Local development
git clone https://github.com/hiroshitanaka-creator/Po_core.git
cd Po_core
pip install -e ".[dev]"

requirements.txt / requirements-dev.txt are repo-local convenience wrappers for a cloned checkout. External consumers should install from package metadata, not from those editable wrappers.

from po_core import run

result = run(user_input="Should AI have rights?")

print(result["status"])       # "ok" or "blocked"
print(result["request_id"])   # Unique request ID
print(result["proposal"])     # Winning philosopher's response

# Optional explicit philosopher allowlist
subset = run(user_input="Should AI have rights?", philosophers=["kant"])

from po_core import PoSelf, PoSelfResponse

po_self = PoSelf(philosophers=["aristotle", "confucius"])
response: PoSelfResponse = po_self.generate("Should AI have rights?")

print(response.text)               # Combined response text
print(response.consensus_leader)   # Winning philosopher name
print(response.philosophers)       # Selected philosopher list
print(response.metrics)            # {"freedom_pressure": ..., "semantic_delta": ..., ...}
print(response.metadata["status"]) # "ok" or "blocked"

# Trace inspection
print(response.log["events"])      # Full trace event stream

# Serialization
d = response.to_dict()
restored = PoSelfResponse.from_dict(d)

export PO_STRUCTURED_OUTPUT=1
python scripts/observe_device.py "転職するべき？家族とキャリアのトレードオフが悩み"

run_ensemble() was removed in v0.3. Use po_core.run() or PoSelf.generate() instead.

po-core version
po-core status
po-core --help

po-self --help
po-trace --help
po-interactive --help
po-experiment --help

Source checkout example:

python examples/po_party_demo.py --help

# Start with recommended auth posture
export PO_API_KEY=dev-secret-key
python -m po_core.app.rest
# → http://localhost:8000  (OpenAPI docs at /docs)

# Reason
curl -X POST http://localhost:8000/v1/reason \
     -H "X-API-Key: dev-secret-key" \
     -H "Content-Type: application/json" \
     -d '{"input": "What is justice?", "philosophers": ["kant"]}'

# Streaming (SSE)
curl -N http://localhost:8000/v1/reason/stream \
     -X POST -H "X-API-Key: dev-secret-key" \
     -H "Content-Type: application/json" \
     -d '{"input": "What is freedom?"}'

# Philosopher manifest
curl -H "X-API-Key: dev-secret-key" http://localhost:8000/v1/philosophers

# Health (no auth required)
curl http://localhost:8000/v1/health

Auth defaults: Keep PO_SKIP_AUTH=false and set a non-empty PO_API_KEY. If both are unset, startup fails fast by design. PO_SKIP_AUTH=true is acceptable only for short-lived local development.

Key env vars (see .env.example for full list):

cp .env.example .env
docker compose up
# API: http://localhost:8000
# Swagger: http://localhost:8000/docs

po-experiment list
po-experiment analyze exp_001_safety_weight_sweep
po-experiment promote exp_001_safety_weight_sweep
po-experiment rollback

Po_core's Pareto weights are config-driven (pareto_table.yaml) and fully externalized. Tune philosophy without code changes; config_version is tracked in all TraceEvents for audit.

We welcome philosophers, engineers, designers, and skeptics.

Contribution tracks:

• AI Track — Start with /04_modules and CLI. Labels: ai-easy, good first issue
• Philosophy Track — Start with /05_research and /glossary. Label: phil-easy
• Bridge Track — Translate checklists to scoring functions. Label: bridge

Flying Pig Philosophy applies: hypothesize boldly, verify rigorously, revise gracefully.

See CONTRIBUTING.md for guidelines.

• "Philosophical Tensor-Based AI Architecture" (in preparation)
• 120+ Technical Specifications in /docs/ and /01_specifications/

@software{po_core2024,
  author = {Flying Pig Philosopher},
  title = {Po_core: Philosophy-Driven AI System},
  year = {2026},
  url = {https://github.com/hiroshitanaka-creator/Po_core}
}

Po_core uses dual licensing:

For commercial licensing: [email protected] See COMMERCIAL_LICENSE.md for details.

Copyright (c) 2024 Flying Pig Project

"If you deny possibilities for pigs, don't eat pork."

Flying Pig Philosopher — Looking up at the sky from the bottom of a well

Built by an independent researcher who asked: "What are AI's possibilities, not its limits?"

• Contact: [email protected]
• Read the full story: Manifesto

• ChatGPT, Gemini, Grok, Claude — My companions throughout this journey
• BUMP OF CHICKEN — For reminding us that even when we say "Leave it to me," we're all a little scared
• Every philosopher who dared to ask "What does it mean to be?"
• You — For believing pigs can fly

The pig has clearance for takeoff.

Po_core: When you must say "Leave it to me," we stand beside you.

"A frog in a well may not know the ocean, but it can know the sky."

⚠️ WARNING: THIS IS THE ORIGINAL Po_core REPOSITORY

• Official sources:

  • GitHub: hiroshitanaka-creator/Po_core
  • Note.com: tensor mania
  • Academia.edu: 僕 僕

• DMCA申請中 (Reference ID: #4124875)

• Any full copy, license rewrite (MIT), or impersonation (flying_pig) will be reported and removed.

• Commercial use requires separate license. Unauthorized copies detected via tensor mania series.