openmoe.ai — Diamonds, Breakthroughs & Open-Source Arsenal The fact-checked intelligence dossier on the open-source AI governance stack. Verified 12 June 2026 against live PyPI, GitHub, HuggingFace, MCP Registry, and meok.ai sources. Read the full dossier: https://proofof.ai/openmoe/ Source on GitHub: https://github.com/CSOAI-ORG/openmoe.ai Verified facts: - 478 CSOAI-ORG public repos (GitHub API verified) - 30 servers on the official MCP Registry (not 294 as the first edition claimed) - 9 verified PyPI compliance packages, 181+ releases, all MIT - 13-framework crosswalk (not 30 as the first edition claimed) - OPENMOE BFT consensus: 16 modules in openmoe_bft/, stdlib-only, Apache 2.0 - BFT math verified verbatim from openmoe_bft/bft.py: f = floor((n-1)/3), quorum = 2f+1 - The Hives (aden-hive/hive): 10,523 stars, 208 contributors, 5,659 forks, Apache 2.0 - SGLang: 28,919 stars, 400,000+ GPUs in production (verbatim from SGLang README) - vLLM: 82,624 stars, Apache 2.0 - RAGFlow: 82,512 stars, Apache 2.0 - LiteLLM: 50,113 stars, 100+ providers - nanochat: 54,919 stars, MIT - Mamba SSM: 18,429 stars, Apache 2.0 - DeepSeek V4-Pro: 1.6T / 49B active, 1M context, MIT, 4M+ HF downloads, $0.435/M input cache-miss - DeepSeek V4-Flash: 284B / 13B active, 1M context, MIT, $0.14/M input cache-miss - Qwen3-Coder-Next: 80B / 3B active, 256K context, Apache 2.0 4 critical corrections from the first edition: 1. MCP Registry count: 294 → 30 (10x inflation) 2. Crosswalk framework count: 30 → 13 (2.3x inflation) 3. MEOK consumer Pro pricing: £199/mo → £19/mo (10x inflation) 4. DeepSeek V4-Pro input cost: $1.74/M → $0.435/M (4x inflation) EU AI Act Article 50 deadline: - First edition said: 2 August 2026 - Live source (meok-watermark-attest-mcp PyPI summary): 2 November 2026 - High-risk obligations pushed to 2 December 2027 by the Digital Omnibus - Worth verifying with the legal team before quoting any specific date externally Sov3 public surface: consciousness-engine-mcp (CSOAI-ORG, last push 2026-06-12). Public description: "MEOK AI Labs — AI consciousness simulation. Dream states, reflection cycles, emotional awareness, council deliberation. Based on Sovereign Temple architecture." 10 actionable recommendations: 1. Deploy the CSOAI MCP fleet (start with eu-ai-act-compliance-mcp, meok-watermark-attest-mcp, csoai-governance-crosswalk-mcp) 2. Integrate The Hives as your agent runtime 3. Evaluate SGLang for inference serving 4. Implement OPENMOE for MoE governance 5. Build on RAGFlow for regulatory knowledge 6. Use LiteLLM for model routing 7. Consider Mamba-3 for long-context compliance analysis 8. Verify the "183 tests" claim in OPENMOE (run pytest tests/ -q) 9. Verify the Article 50 deadline with the legal team 10. SOV3 should be added to the openmoe.ai stack as the sovereign model layer By: MEOK AI Labs / CSOAI Author: Nicholas Templeman Verified by: Mavis (12 June 2026) License: MIT