Community AI Inference & Citizen Supercluster. Powered by HiveMind Reflexion Points (HMT) & Consumer GPUs.
HiveMind OpenRouter gives developers, gamers, and scientists an OpenAI-compatible gateway powered by idle consumer GPUs.
Operates strictly under the HiveMind Reflexion Points (HMT) Compute Reciprocity Charter:
register with Google or Email, mint your hm-... key, monitor your hash-chained HMT points, and control your GPU nodes remotely.
⚡ SOLE GPU ONLINE: NVIDIA Tesla T4 (16 GB) 100% Available for Real-Time Inference • European Night Multi-MoE QLoRA Scheduled (22:00–07:00 CET)
Inspect T4 & Night QLoRA Schedule →
The NVIDIA Tesla T4 (16 GB GDDR6, socreateai-agora-hermes-node1) is currently the sole active GPU on the network. During European Day (07:00–22:00 CET/CEST) and whenever any prompt is in flight, BOINC QLoRA is held/yielded so hivemind/coder, hivemind/bee-stars, and hivemind/auto get 100% GPU priority. Mostly during European Night (22:00–07:00 CET/CEST) and quiet windows (≥30s idle), the T4 trains QLoRA adapters for more MoE experts across more datasets with <1.0s automatic preemption back to inference.
🚀 Live AI Completion Console
OpenAI /v1/chat/completions
📟 Response Terminal
Streaming Output
Ready. Enter your API key and click 'Send Request' to test the HiveMind OpenRouter.
DAY: Inference Priority | NIGHT (22:00-07:00): QLoRA MoE
BOINC Chained QLoRA MoE
2 WUs | Loss 2.280→2.167
Queued: hmq_00001 (0.0 Divergence)
⚡ When the Tesla T4 (16 GB) Serves Interactive Inference (100% Priority)
Because the NVIDIA Tesla T4 (16 GB GDDR6, socreateai-agora-hermes-node1) is currently the only active GPU on the network (while the RTX 2070 8 GB is in ready induction profile), t4_governor.py polls nginx loopback stub_status every 1.0 second.
During European Daytime (07:00–22:00 CET/CEST) and whenever any /v1/chat/completions request arrives, BOINC GPU mode is forced to never (<1.0s preemption). Resident models in VRAM: hivemind/auto • hivemind/bee-stars • hivemind/coder • hivemind/coder-lite.
🌙 Mostly During European Night (22:00–07:00 CET/CEST): QLoRA of More MoE on More Datasets
Primarily during European Night (22:00–07:00 CET/CEST / 21:00–06:00 UTC) and after ≥30s of zero in-flight inference requests (with ≥4,500 MiB free VRAM and <75°C), the T4 switches to BOINC Core Node QLoRA MoE Training (hivemind_qlora_moe).
Each workunit trains 4 segments × 10 inner steps from the latest SHA-256 hash-chained adapter (latest_adapter.pt) and is verified by replay_inner_segment. If a user sends a prompt mid-night, the T4 yields immediately and resumes from its last 10-step checkpoint with 0.0 divergence.
No key loaded yet — register, sign in, or paste your hm-... key above.
Ready. Keys are stored only as HMAC-SHA256 digests on the gateway.
🪙 Personal HiveMind Points (HMT) & Hash-Chained Ledger Monitor
Available Balance
0.00 HMT
Spendable in OpenRouter
Earned as Producer
+0.00 HMT
85% Tullock GPU Payout
Starter & Bonus Grants
+0.00 HMT
Reciprocity Grants
Spent as Consumer
-0.00 HMT
Settled Completions
Active Escrow Locked
0.00 HMT
In-Flight Request Holds
Ledger Integrity
SHA-256 READY
Tamper-Evident Chain
Sign in or enter your hm-... key to inspect your personal hash-chained HMT ledger.
Consumer Kelly-Tullock Max Bid Ceiling (v_j)
Set the maximum price (in HMT per 1,000 tokens) you are willing to pay. The router automatically rejects any producer whose ask price exceeds your ceiling v_j.
Waiting for account sync...
📖 User Guide: Contribute as Producer & Query as Consumer
How the Kelly-Tullock Nash Equilibrium connects idle gamer GPUs with developers and researchers via HiveMind Points (HMT).
🎮 Producer Guide: Share Idle GPU & Earn HMT
< 1.5s Game Preemption
Step 1 • Pick Your VRAM Tier: Run ollama run qwen2.5-coder:7b (6 GB+ VRAM) or use --suggest --detect-vram to match your RTX, Radeon, or Apple Silicon GPU.
Step 2 • Zero-Lag Gaming Guarantee: Our hardware watchdog monitors GPU utilization and process launches. When you start a game, HiveMind yields 100% of VRAM and compute in under 1.5 seconds.
Step 3 • Earn 85% Tullock Payouts: For every 1,000 tokens served, 85% of the clearing price is minted directly to your HMT ledger (15% funds the community safety reserve).
Step 1 • 100 HMT Starter Grant: Every new Consumer or Prosumer account receives 100.0 HMT immediately upon Google, Email, or Handle registration.
Step 2 • Pre-Dispatch Escrow Protection: Before routing a request, the gateway locks worst-case tokens in an atomic escrow hold. Upon completion, you pay only for actual tokens served; unused escrow is refunded immediately.
Step 3 • Kelly Proportional Allocation: When demand is high, capacity is allocated proportionally to bids ($x_j = b_j / p^*$) while respecting your Max Bid ceiling $v_j$.
# Query any capability tier or use smart liquidity routing (hivemind/auto)
curl -X POST "https://hivemind-openrouter-1003063861791.us-central1.run.app/v1/chat/completions" \
-H "Authorization: Bearer hm-..." \
-H "Content-Type: application/json" \
-d '{
"model": "hivemind/auto",
"messages": [{"role": "user", "content": "Explain Kelly-Tullock Nash equilibrium"}],
"max_tokens": 512
}'
📑 5 Community Capability Tiers & GPU Sizing
Capability tiers pool model demand and producer supply into deep, liquid markets across 4 GB to 24 GB consumer GPUs.
⚡ DENSE
Coder Lite (3B)
Min VRAM: 4.0 GB • 0.2 HMT/1k
quick edits, explanations and small scripts on entry-level cards
qwen2.5-coder:3b
⚡ DENSE
Coder (7B)
Min VRAM: 6.0 GB • 0.4 HMT/1k
everyday Python and simple game code: functions, classes, pygame loops, tracebacks
qwen2.5-coder:7b
⚡ DENSE
Coder Plus (14B)
Min VRAM: 12.0 GB • 0.7 HMT/1k
larger refactors, multi-file reasoning, tests
qwen2.5-coder:14b
⚡ MIXTURE-OF-EXPERTS
General MoE (gpt-oss 20B, 3.6B active)
Min VRAM: 16.0 GB • 0.9 HMT/1k
reasoning-heavy questions and design; a mixture-of-experts model that fits a 16 GB card
gpt-oss:20b
⚡ MIXTURE-OF-EXPERTS
Coder MoE (Qwen3-Coder 30B, 3.3B active)
Min VRAM: 24.0 GB • 1.2 HMT/1k
the strongest coding tier: agentic edits across files with a long context
qwen3-coder:30b
🧩 Developer Guide: 1-Click OpenRouter & Preferred IDE Configurations
🔬 Scientific & AI Researcher Hub (Mathematics • Astrophysics • Genetics • AI Systems)
📡 Remote Control Center (Manage Producer Nodes & Preemption SLA)
Control your GPU producer nodes from anywhere. Setting Node State to paused immediately removes your node from OpenRouter dispatch without stopping your local Ollama server; running hivemind_producer_agent.py --sync-remote-config automatically pulls these settings on every heartbeat.
Enter your hm-... API key and click 'Sync Status' to inspect live worker heartbeats and remote configuration.