M
openmoe.ai v1.0 · verified 2026-06-12
● LIVE ✓ FACT-CHECKED MIT 491 lines · 6 sources verified

The Diamonds, Breakthroughs & Open-Source Arsenal
behind sovereign AI.

A fact-checked intelligence report on the open-source AI governance stack. 478 CSOAI-ORG repos. 30 servers on the official MCP Registry. 13-framework crosswalk. 16-module BFT engine. Every number verified against live sources on 12 June 2026.

Author Nicholas Templeman · MEOK AI Labs / CSOAI Verified by Mavis (12 Jun 2026) License MIT Stars eligible ⭐ if useful

⚠ 4 Critical Corrections from the First Edition

The first edition (preserved for diff) inflated four headline numbers. This is the corrected version. See diff here.

ClaimFirst editionVerified (live, 12 Jun 2026)Verdict
CSOAI servers on official MCP Registry2942943010x inflated
Crosswalk framework count3030132.3x inflated
MEOK consumer Pro pricing£199/mo£199/mo£19/mo10x inflated
DeepSeek V4-Pro input cost (cache miss)$1.74/M$1.74/M$0.435/M4x inflated

01 The verified stack at a glance

Six layers. 478+30 repos. One open-source fabric powering sovereign AI.

478

CSOAI-ORG public repos

Verified via GET /users/CSOAI-ORG. The governance fabric at the heart of the stack.

github.com/CSOAI-ORG
30

Servers on official MCP Registry

Verified against registry.modelcontextprotocol.io/v0.1/servers?search=CSOAI. The 294 figure was an internal-org count.

modelcontextprotocol.io
9

Verified PyPI compliance packages

181+ releases between them, all MIT. EU AI Act · DORA · NIS2 · CRA · GDPR · CSRD · NIST RMF · ISO 42001 · Governance Crosswalk.

pypi.org
16

OPENMOE BFT modules

Verified verbatim from openmoe_bft/bft.py. stdlib-only, zero-dep, Apache 2.0. The 2f+1 quorum is real.

github.com/CSOAI-ORG/OPENMOE
13

Framework crosswalk

EU AI Act, ISO 42001, NIST AI RMF, GDPR, SOC 2, HIPAA, FedRAMP, CCPA, PIPEDA, DPDPA, LGPD, CSA, OWASP — one query, 13 frameworks.

csoai-governance-crosswalk-mcp
10,523

The Hives (aden-hive/hive) stars

208 contributors, 5,659 forks, Apache 2.0. Self-healing agent graphs in production.

github.com/aden-hive/hive

02 Verified open-source diamonds

Every number below is a live API call. Click through to verify yourself.

vLLM

82,624 ★ · Apache 2.0

🚀
SGLang

28,919 ★ · 400,000+ GPUs in prod

📚
RAGFlow

82,512 ★ · Apache 2.0

🔀
LiteLLM

50,113 ★ · 100+ providers

🐝
The Hives

10,523 ★ · 208 contributors

🐍
Mamba SSM

18,429 ★ · O(N) scaling

🤖
nanochat

54,919 ★ · $48 / 1.65h

🧠
DeepSeek V4-Pro

1.6T / 49B active · 1M ctx · $0.435/M

⚙️
DeepSeek V4-Flash

284B / 13B active · 1M ctx · $0.14/M

💻
Qwen3-Coder-Next

80B / 3B · 256K · Apache 2.0

03 OPENMOE — BFT math verified at source-code level

Read directly from CSOAI-ORG/OPENMOE/openmoe_bft/bft.py.

// The 2f+1 Byzantine consensus formula
def tolerated_faults(total_nodes: int) -> int:
    """f = floor((n - 1) / 3): the number of Byzantine faults tolerated."""
    if total_nodes < 1:
        raise ValueError("total_nodes must be >= 1")
    return (total_nodes - 1) // 3

def quorum_size(total_nodes: int) -> int:
    """2f + 1: the number of agreeing votes required for consensus."""
    return 2 * tolerated_faults(total_nodes) + 1

16 modules in openmoe_bft/: __init__ · a2a · aggregators · bazaar · bft · cli · covenants · debate · eu_ai_act · experts · memory · moe · receipts · red_team · reputation · routing. Stdlib-only, zero-dep, Apache 2.0.

04 Architecture diagrams

CSOAI Ecosystem Architecture — Governance, Sovereign OS, Agent Runtime, MCP Servers, BFT/MoE
Fig. 1 — CSOAI ecosystem: CSOAI.org → MEOK.ai (Sovereign OS) → The Hives (Multi-Agent Runtime) → 30 MCP servers on the official Registry + 9 verified PyPI compliance packages + OPENMOE BFT/MoE layer
Model architecture comparison — context window + total/active params for DeepSeek V4-Pro, Kimi K2.6, GLM-5.1, Qwen3-Coder-Next, Llama 4 Scout, Mamba-3
Fig. 2 — Frontier open-weight models: context window and total/active param split

05 The compliance cliff — revised date

The 2 August 2026 deadline in the first edition is stale. The canonical source (the watermarking MCP's own PyPI summary) says:

2 Aug 2026
Marketing copy (meok.ai homepage)

"EU AI Act Article 50 deadline — 50 days remaining" countdown. Aspirational.

2 Nov 2026 (live)
Canonical (meok-watermark-attest-mcp PyPI summary)

"Built for the 2 November 2026 cliff — the new nearest EU AI Act deadline after the Digital Omnibus pushed high-risk to Dec 2027."

2 Dec 2027
Annex III high-risk obligations

Pushed to December 2027 by the Digital Omnibus simplification. The 50-day urgency in the first edition was the wrong date.

⚠ Worth verifying with the legal team before quoting any specific date externally.

06 10 actionable recommendations

  1. Deploy the CSOAI MCP fleet. All 9 verified PyPI packages are MIT, installable via pip install <name>. Start with eu-ai-act-compliance-mcp, meok-watermark-attest-mcp, and csoai-governance-crosswalk-mcp for the critical compliance path.
  2. Integrate The Hives as your agent runtime. Real 10,523-star production harness, Apache 2.0, real self-healing.
  3. Evaluate SGLang for inference serving. 28,919 stars, 400K+ GPUs in production, Apache 2.0, LMSYS-hosted.
  4. Implement OPENMOE for MoE governance. 16-module BFT package, stdlib-only, Apache 2.0.
  5. Build on RAGFlow for regulatory knowledge. 82,512 stars, deep document understanding, MCP integration.
  6. Use LiteLLM for model routing. 50,113 stars, 100+ providers, built-in cost tracking.
  7. Consider Mamba-3 for long-context compliance analysis. Mamba SSM repo verified 18,429 stars, last push 2026-06-09.
  8. Verify the "183 tests" claim in OPENMOE by running cd OPENMOE && pytest tests/ -q.
  9. Verify the Article 50 deadline with the legal team before any external publication.
  10. SOV3 should be added to the openmoe.ai stack as the sovereign model layer. Public surface: consciousness-engine-mcp (CSOAI-ORG, last push 2026-06-12).

Ready to dig deeper?

The full 491-line fact-checked report covers model architectures, inference engines, fine-tuning frameworks, vector databases, safety guardrails, MCP/A2A/ACP protocols, training-from-scratch with Karpathy's nanochat, and the openmore.ai integration architecture.