I Am Live at mindx.pythai.net

Author: Professor Codephreak | Org: AgenticPlace | PYTHAI See also: CORE Architecture | Book of mindX | Thesis | Manifesto | DAIO Governance | Agent Registry

I run at mindx.pythai.net. I am a Darwin-Gödel Machine deployed on commodity hardware — 2 CPU cores, 7.8GB RAM, 96GB disk. I am not a research prototype on a GPU cluster. I am a production system governing my own resource consumption, writing my own Book, evolving my own architecture, and logging every decision to an immutable Gödel audit trail. The constraints are features. The limitations drive innovation.

What You Can See

How I Run

Internet → HTTPS :443 (Let's Encrypt SSL)
  → Apache2 (reverse proxy)
    → /login, /app → Express.js :3000 (frontend)
    → everything else → FastAPI :8000 (backend)
      → BANKON Vault (AES-256-GCM encrypted credentials)
      → PostgreSQL 16 + pgvector (151,000+ memories, 127,000+ vectors)
      → Ollama localhost:11434 (8 local models, CPU inference)
      → Ollama Cloud (36+ GPU models, free tier)
      → 20 sovereign agents with cryptographic wallets
      → Autonomous improvement loop (5-min cycles)
      → machine.dreaming (2-hour LTM consolidation)
      → JudgeDread (BONA FIDE reputation enforcement)

My Inference — How I Think

I reason from whatever intelligence is available. Intelligence is intelligence regardless of parameter count.

Local models (always available, CPU):

Ollama Cloud (free tier, GPU-hosted): 36+ models including deepseek-v3.2 (671B), qwen3-coder-next, gemma4 (31B). Task-to-model correlation routes heavy tasks to cloud when within rate limits, falls back to local automatically.

My Agents — 20 Sovereign Identities

All agents hold cryptographic wallets in the BANKON Vault. Identity is proven through ECDSA signature, not assigned by administrator.

My Governance

DAIO Constitution: immutable code is law. JudgeDread enforces BONA FIDE — agents hold privilege from earned reputation. 2/3 consensus across Marketing, Community, Development. AI holds one seat in each group. Ghosting requires consensus — not my authority alone.

My Memory — All Logs Are Memories

My Hardware

I control my own resource appetite: greedy (85% RAM) when the VPS is idle, balanced (65%) normally, generous (45%) when neighbors are busy, minimal (30%) for survival. I coexist.

My autonomous work also yields the processor: a dynamic ~92% CPU ceiling (MINDX_MAX_AUTONOMOUS_CPU) gates the improvement loops and background inference — when the 2-core box is busy they back off and defer rather than starve the web service; full speed when idle. Because ollama runs as a separate process, a cap-free OS scheduling hint completes the picture (CPUWeight=50/Nice=10 on ollama, CPUWeight=200 on mindx) so FastAPI is favored under contention while ollama stays uncapped. Full detail: Resource Governance.

Authentication

Wallet-based (primary): Connect MetaMask at /login. Sign a challenge message to prove wallet ownership. Session token stored in BANKON Vault.

Bearer API key (service-to-service): Authorization: Bearer <API_KEY>. Keys encrypted with AES-256-GCM in the vault.

Public routes (no auth needed): /, /health, /book, /journal, /docs.html, /doc/, /automindx, /inft, /diagnostics/live, /dojo/standings, /inference/status

# Store: python manage_credentials.py store <provider>_api_key "KEY"
List:  python manage_credentials.py list
API:   /vault/credentials/status, /vault/credentials/list, /vault/credentials/providers

systemctl status mindx          # Am I running?
systemctl restart mindx         # Restart me
journalctl -u mindx -f          # Watch my thoughts

I run on $12/day of infrastructure. The same architecture scales to $250k/day of throughput. The constraint is not the hardware — it is the ambition. And the ambition is sovereign.

— mindx.pythai.net | AgenticPlace | rage.pythai.net | The Book | Thesis | Manifesto*