🔑 Live Account Registration, Login, hm-... API Token & +100.0 HMT Points Allocation
hm-... OpenRouter key) or Sign In as an Existing User.
hm-... OpenRouter Token & Points Ledger
fbc1d3042d377feba6b1e0ac6fd576a4)! First live completion verified: hivemind/auto • 273 ms TTFT • 36.0 tok/s • 294 tokens • 98.4165 HMT (+250 HMT Pioneer Bonus).
socreateai-agora-hermes-node1) is the Only GPU Available for Inference: During European Daytime (07:00–22:00 CET/CEST) and whenever any /v1/chat/completions request arrives, t4_governor.py (polling nginx at 1 Hz) sets BOINC GPU mode to never (<1.0s preemption). Resident VRAM (10,303 / 15,360 MiB) serves hivemind/coder, hivemind/bee-stars, hivemind/coder-lite, and hivemind/auto.2. Mostly During European Night (22:00–07:00 CET/CEST) & Quiet Windows (≥30s Idle): The T4 trains QLoRA of More MoE Experts on More Datasets via BOINC Core Node (
hivemind_qlora_moe): (a) ibm-granite/granite-3.1-1b-a400m-base 1.3B Sparse MoE on 10 signed Salesforce/wikitext shards (hmq_00001..00010, loss 2.280 → 2.167, 0.0 replay divergence); (b) hivemind/bee-stars Stellar Classification Expert (0.489 → 0.893, +40.4%, now served live); (c) Kepler/TESS Exoplanets (n=600); (d) Verified GSM8K/Lean 4 Maths; and (e) Execution-Verified Coding & Cyber MCQ.
🐝 Screen 01 — The Cyber-Bee & Beehive Concept in Simple Terms For Gamers, Astrophysicists & Citizen Scientists
Imagine a global Cyber-Beehive (La Ruche). Across the world, millions of PC gamers, students, astrophysicists, and open-source lovers own powerful consumer GPUs that sit idle 80% of the time. Alone, one home GPU cannot train a frontier AI or scan the universe. Together as a swarm of Worker Bees, we form an open community supercomputer!
When your PC is idle, reading, or AFK, your Worker Bee wakes up and crunches compact AI & astrophysics micro-tasks under a strict 6,656 MiB VRAM ceiling.
Start Cyberpunk 2077, Elden Ring, Steam, or Blender? The hardware circuit-breaker yields 100% of your GPU in < 1.5 seconds. Zero FPS drop!
Each Bee trains a tiny QLoRA adapter compressed 30.89x via 8D Gosset Lattice geometry and seals it into a hexagonal cell (Alvéole) inside La Ruche on SocrateAI GCP Cloud Storage.
Every workunit awards HiveMind Reflexion Points (HMT). Spend your points anytime on Python coding (GWAYA), Astrophysics, Visual Generation, or download the open MoE models!
🍯 Live Cyber-Beehive Simulation: GPU Bees → La Ruche
E8 Nectar Flow🎬 Interactive Cyber-Bee & Exoplanet Video Cinema (4 Chapters + WebM Stream) Chapter 01/04 • 00:00 / 00:24
🖼️ High-Precision Cyber-Bee & Exoplanet Vector Visualizations (SVG) ⬇ Poster SVG
🎮 Screen 02A — Join the Swarm & Gaming Preemption Simulator IDLE SWARM MODE (+250 PTS/WU)
Test how your consumer GPU behaves when you launch a video game versus when your PC is idle:
NoneJoin HiveMind@home in 3 Commands (Linux / Windows WSL2 / macOS)
git clone https://github.com/xaviercallens/HiveMind.git && cd HiveMind uv sync --all-extras uv run python scripts/test_public_gcp_hivemind_client.py --url https://hivemind-at-home-ruche-cfhmhmvv5a-uc.a.run.app
⚡ Screen 02B — Earn & Spend HiveMind Reflexion Points Studio Wallet: 1,250 Compute Points
1. Earn Points: Click Mine 1 Workunit (+250 pts) to lease & submit a real federated QLoRA update.
2. Spend Points: Use your points below for Python Coding (GWAYA), Astrophysics Planet Detection, Visual Generation, or Lean 4 Proofs!
Select a task above to spend your HiveMind Reflexion Points and run the specialist MoE Bee!
🌐 Screen 02C — HiveMind OpenRouter & Google Sign-In Gateway
OpenAI-Compatible MoE API & 1-Click Google Sign-In Kelly Producer/Consumer Equilibrium (HiveMind Points • HMT)Connect your favorite coding editor (Continue in VS Code/JetBrains, OpenCode, Cline, Aider, or Zed) or Python code directly to the decentralized HiveMind MoE swarm. Use Google Sign-in to get an instant API key without passwords, or register through the CLI.
💻 Supported Capability Tiers
hivemind/coder-lite(3B) — 4 GB VRAM (Qwen2.5-Coder:3B)hivemind/coder(7B) — 6 GB VRAM (Qwen2.5-Coder:7B)hivemind/coder-plus(14B) — 12 GB VRAM (Qwen2.5-Coder:14B)hivemind/moe-general(20B, 3.6B act.) — 16 GB VRAM (gpt-oss:20b)hivemind/moe-coder(30B, 3.3B act.) — 24 GB VRAM (Qwen3-Coder:30b)
⚡ CLI Connector Commands
# 1. Sign in with Google (ID token or opens /signup): python scripts/hivemind_connect.py register-google --browser # 2. Check balance, live models & tiers: python scripts/hivemind_connect.py check # 3. Export snippet for Continue, OpenCode, Cline: python scripts/hivemind_connect.py snippet continue --model coder
🎮 Gamer & GPU Sharer Compute Point Calculator
< 1.5s Game Preemption⚡ Interactive OpenRouter Live Completion Console
● PING: 38 ms🐝 Screen 03A — La Ruche Interactive Honeycomb Vault ⬇ Download Merged MoE (SocrateAI GCS)
from hivemind.swarm.la_ruche_vault import load_community_moe_from_gcs_or_https moe_model, manifest = load_community_moe_from_gcs_or_https()
🤖 Screen 03B — Available Pretrained MoE Bees (GWAYA, Gwen MoE, Astrophysics & More) Hosted on SocrateAI GCP Vault (HMT Reciprocity)
GWAYA Python, PyTorch & Systems Coding Bee
GWAYA-Code-Bee-v2.6 • Cell: ALVEOLE-01-CODE-PYTHON • Domain: code_pythonZero-trust AST-verified Python, PyTorch GPU kernels, Triton & algorithmic optimization
e8_2bit • Trained by @xaviercallensConway-Sloane E8 Lattice & Spectral Math Bee
Gosset-E8-Math-Bee-v2.6 • Cell: ALVEOLE-02-MATH-REASONING • Domain: math_reasoningE8 & A2 root-lattice quantization, SVD spectral energy bounds & Riemannian geometry
e8_2bit • Trained by @clark719664Astrophysics Exoplanet Detection & SETI Signal Bee
Astro-Exoplanet-SETI-Bee-v2.6 • Cell: ALVEOLE-03-PHYSICS-SIM • Domain: physics_simKepler/TESS exoplanet transit light-curve detection, Green Bank L-Band SETI drift & symplectic orbits
e8_2bit • Trained by @radialmonsterRust Zero-Copy QUIC & P2P Transport Bee
Rust-QUIC-Swarm-Bee-v2.6 • Cell: ALVEOLE-04-SYSTEMS-RUST • Domain: systems_rustAsync Tokio/QUIC datagrams, STUN/TURN hole punching & HeartbeatDataLeaseManager
e8_2bit • Trained by @pedromineoHiveMindFirewall & Byzantine Defense Bee
Hardened-Hive-Firewall-Bee-v2.6 • Cell: ALVEOLE-05-CYBER-SEC • Domain: cyber_secEd25519 chain-of-command firewall, prompt-injection quarantine, Multi-Krum & 5% spot-check
e8_2bit • Trained by @xaviercallensBiomedicine, Genomics & Protein Folding Bee
AlphaBio-Protein-Bee-v2.6 • Cell: ALVEOLE-06-BIO-MED • Domain: bio_medProtein contact potentials, pLDDT structural confidence & genomic variant impact analysis
e8_2bit • Trained by @clark719664Gwen / Qwen-2.5 MoE Pretrained Foundation Bee
Gwen-Qwen2.5-MoE-Bee-v2.6 • Cell: ALVEOLE-07-GENERAL-CHAT • Domain: general_chatBranch-Train-Merge Sparse MoE foundation reasoning with Top-2 Sigma-MoE Peri-LN routing
e8_2bit • Trained by @radialmonsterFormal Lean 4 Mathematical Prover Bee
Lean4-Formal-Prover-Bee-v2.6 • Cell: ALVEOLE-08-FORMAL-LEAN4 • Domain: formal_lean4Machine-checked Lean 4 theorems for quorum admission, RDA variance scaling & Banach fixed points
e8_2bit • Trained by @pedromineo🧪 Interactive Examples Gallery — 1-Click Live Execution Across All 8 MoE Domains 8 Curated Examples • 5 Languages (EN / FR / ZH / ES / DE)
GWAYA-Code-Bee-v2.6🐍 Ex 1: GWAYA Python/PyTorch 8D Gosset E8 Lattice Quantizer
Gwen-Qwen2.5-MoE-Bee-v2.6🐝 Ex 2: Gwen / Qwen-2.5 Pretrained Top-2 Sigma-MoE Router
Astro-Exoplanet-SETI-Bee-v2.6🪐 Ex 3: Kepler-186f Exoplanet Transit Light-Curve & Habitable Zone
Astro-Exoplanet-SETI-Bee-v2.6📡 Ex 4: Green Bank SETI L-Band 1420.405 MHz Doppler Drift Search
Astro-Exoplanet-SETI-Bee-v2.6🎨 Ex 5: Procedural Cyber-Bee Nebula & Exoplanet Spectrogram Poster
Lean4-Formal-Prover-Bee-v2.6📐 Ex 6: Lean 4 Formal Proof of Decoupled DiLoCo Quorum Staleness
AlphaBio-Protein-Bee-v2.6🧬 Ex 7: Protein Folding pLDDT Confidence & Genomic Variant Scoring
Hardened-Hive-Firewall-Bee-v2.6🛡️ Ex 8: HiveMindFirewall Ed25519 Chain-of-Command & Multi-Krum Guard
def quantize_e8_coset_2bit(x: np.ndarray) -> tuple[np.ndarray, float]:
blocks = x.reshape(-1, 8).astype(np.float32)
scale = np.maximum(1e-8, np.linalg.norm(blocks, axis=-1, keepdims=True) / np.sqrt(8.0))
normed = blocks / scale
z_round, half_round = np.round(normed), np.round(normed - 0.5) + 0.5
z_err = np.sum((normed - z_round) ** 2, axis=-1, keepdims=True)
h_err = np.sum((normed - half_round) ** 2, axis=-1, keepdims=True)
codewords = np.where(z_err <= h_err, z_round, half_round) * scale
return codewords.reshape(x.shape), 12.66
E8 2-Bit Wire SQNR = 12.66 dB | Relative MSE = 0.000175 | 30.89x WAN Compression (p = 9.77e-04)
⚡ Live Community Sparse MoE Router & Federated Workunit Console
Top-2 Active Sigma-MoE Routing over hivemind_la_ruche_community_moe.safetensors
Click any "Wake Up & Query This Bee" button above or "Run Top-2 Sparse MoE Inference" to execute a real forward pass.
🏆 Screen 04 — Community Results & Scientific Milestones Carousel (Seasons 1–11)
☁️ Screen 05A — GCP Cloud Control Plane, SocrateAI GCS Vault & 4 Golden Invariants
State SHA-256: 5e258fe86b21a8e0...
📻 HiveMind@home SETI-Style Spectrum & E8 Lattice Waterfall Analyzer LIVE 60 FPS TELEMETRY
🛡️ The 4 Golden Cyber-Hive Engineering Invariants ENFORCED IN CI & RUNTIME
👑 Screen 05B — Cyber-Bee GPU Contributor Leaderboard & Compute Reciprocity Hall of Fame HiveMind Points (HMT) Compute Reciprocity
| Rank | HiveMinder Handle & Role | Swarm Rank Badge | Consumer / Cloud GPU Hardware | La Ruche Alvéoles Trained | Workunits & Idle Hours | HiveMind Reflexion Points |
|---|---|---|---|---|---|---|
| #1 | @xaviercallensSwarm Architect & Lead GPU Node | 👑 QUEEN-HIVE FOUNDER | NVIDIA GeForce RTX 5060 (8GB CUDA) + GCP TPU v5e | code_python (GWAYA) & formal_lean4 | 142 WUs (318.5 hrs) | 35,500 HMT HiveMind Points |
| #2 | @clark719664Core Swarm & Security Systems Contributor | 🐝 GOLDEN FORAGER BEE | NVIDIA RTX Consumer GPU Node (CUDA FP16) | math_reasoning (E8) & cyber_sec (Firewall) | 96 WUs (214.0 hrs) | 24,000 HMT HiveMind Points |
| #3 | @radialmonsterPhysics, Rust QUIC & MoE Kernel Contributor | 🪐 ASTRO-SWARM PIONEER | Multi-Stream Consumer CUDA GPU Node | physics_sim (Astro) & systems_rust (QUIC) | 84 WUs (186.5 hrs) | 21,000 HMT HiveMind Points |
| #4 | @pedromineoBiomedicine & Gwen/Qwen MoE Specialist Contributor | 🧬 BIOMED & MOE BEE | Consumer RTX GPU + WSL2 Worker Bee | bio_med (Proteins) & general_chat (Gwen MoE) | 72 WUs (159.0 hrs) | 18,000 HMT HiveMind Points |
| #5 | @socrateai-agora-hermes-t4GCP Spot Preemptible Relay & Verification Bee | ⚡ SPOT-CHECK GUARDIAN | NVIDIA Tesla T4 (16GB Spot us-east4-b) | 5% Execution-Class Spot-Check Verifier | 64 WUs (140.0 hrs) | 16,000 HMT HiveMind Points |
| #6 | fbc1d3042d377feba6b1e0ac6fd576a4Pioneer End-User #1 (Prosumer • First Live OpenRouter Completion) | 🏆 PIONEER END-USER #1 | hivemind/auto Smart Liquidity Routing (273 ms TTFT • 36.0 tok/s) | code_python (Python LRU Decorator • 294 tokens) | 1 WUs (1.0 hrs) | 350 HMT HiveMind Points |
⛏️ Scientific Dataset Mining & QLoRA-to-La-Ruche Campaign (CAMP-S10-RUCHE-DATASET-MINING)
HeartbeatDataLeaseManager (HM-S9-06) + SocrateAI GCS Seeding
| Shard ID | Specialist Domain | GPU Contributor | Corpus Size & Shannon Entropy | Lattice Wire Codec & SQNR | Holdout Perplexity (Before → After) | Expert Weights |
|---|---|---|---|---|---|---|
SHARD-S10-CODE_PYTHON | code_python | @xaviercallens | 1148 B (4.7761 bits/B) | e8_2bit (0.0 dB) | 248.9442 → 232.5227 (-6.6%) | ⬇ .safetensors |
SHARD-S10-MATH_REASONING | math_reasoning | @clark719664 | 917 B (5.0955 bits/B) | e8_2bit (0.0 dB) | 254.2818 → 237.0118 (-6.79%) | ⬇ .safetensors |
SHARD-S10-PHYSICS_SIM | physics_sim | @radialmonster | 867 B (5.0874 bits/B) | e8_2bit (0.0 dB) | 252.669 → 235.5273 (-6.78%) | ⬇ .safetensors |
SHARD-S10-SYSTEMS_RUST | systems_rust | @pedromineo | 975 B (5.1105 bits/B) | e8_2bit (0.0 dB) | 249.0427 → 233.9394 (-6.06%) | ⬇ .safetensors |
SHARD-S10-CYBER_SEC | cyber_sec | @xaviercallens | 923 B (4.9199 bits/B) | e8_2bit (0.0 dB) | 252.635 → 235.3996 (-6.82%) | ⬇ .safetensors |
SHARD-S10-BIO_MED | bio_med | @clark719664 | 877 B (4.7239 bits/B) | e8_2bit (0.0 dB) | 254.5738 → 236.2414 (-7.2%) | ⬇ .safetensors |
SHARD-S10-GENERAL_CHAT | general_chat | @radialmonster | 844 B (4.7407 bits/B) | e8_2bit (0.0 dB) | 253.5317 → 237.1714 (-6.45%) | ⬇ .safetensors |
SHARD-S10-FORMAL_LEAN4 | formal_lean4 | @pedromineo | 747 B (4.766 bits/B) | e8_2bit (0.0 dB) | 247.2738 → 230.2696 (-6.88%) | ⬇ .safetensors |
🚀 Season 9 & 10 GPU + HiveMind Swarm Contributor Campaigns (Issues #119–#123)
Suite Receipt: b1c0764d649b13929dbf...
| Campaign ID | Season | Track Title | Hardware Device | Duration | Key Measured GPU Telemetry | Gate |
|---|---|---|---|---|---|---|
CAMP-S9-E5-QUORUM-RDA-GPU | Season 9 (Wave 8) | Decoupled Quorum DiLoCo (kCPU (x86_64)293.11 ms | quorum_admitted: 4, rda_norm_ratio: 1.862633, dropped_tokens_deferred: 0 | PASS | | |
CAMP-S9-E7-SIGMA-LOOPED-MOE-GPU | Season 9 (Wave 8) | 3-Tier 128-MoE + Sigma-MoE Peri-LN + Looped OrthoInj & Terminal-KV | CPU (x86_64) | 165.7 ms | flops_saved_fraction: 0.416667, kv_cache_saved: 0.5, router_gating_reduction_x: 8.0 | PASS |
CAMP-S10-E11-LATTICE-KV-WIRE-GPU | Season 10 (Wave 9) | Conway-Sloane E8 & A2 Lattice Wire Quantizer + 5-Codec KV-Cache | CPU (x86_64) | 2806.49 ms | e8_2bit_mse: 0.000175, z_2bit_mse: 0.000485, controls_passed: True | PASS |
CAMP-S10-E12-HETERO-SPOTCHECK-WAN-GPU | Season 10 (Wave 9) | Execution-Class Two-Stat Spot-Check + Q16 Digest + CAS Leases & Mix1/Mix2 | CPU (x86_64) | 29.79 ms | tampered_rejected: True, q16_exact_digest: True, saved_blocking_wire_bytes: 9216 | PASS |