# ElephantPool

> ElephantPool is a distributed AI inference network by De Buck Technologies (Belgium). Developers call an OpenAI-compatible API to run open-source models (Llama, Qwen, DeepSeek, gpt-oss); requests are served by a verified pool of GPUs across Europe — from gaming rigs to datacenter nodes. Hardware owners install the open-source `mahout` agent and are paid 80% of the revenue of every token their machines serve. Monthly SEPA payouts via Stripe.

ElephantPool is **not** an investment product. Hosts are paid service fees for compute work actually performed. Clients pay per token, prepaid, no subscription.

## For builders (API clients)

- OpenAI-compatible endpoints (`/v1/chat/completions`, streaming) — change one base URL.
- Latest open-weight models only. Launch catalog: Llama 3.1 8B, Qwen3.8 27B, Gemma 4 26B-A4B, Llama 3.3 70B, gpt-oss-120b, DeepSeek V4-Flash (304B MoE), MiniMax M3 (427B MoE), GLM-5.2 (753B MoE), Qwen3.5 397B-A17B.
- Target launch pricing (€/M tokens in · out, standard tier): Llama 8B €0.03·€0.06, Qwen3.8 27B €0.25·€1.50, Gemma 4 26B-A4B €0.06·€0.25, Llama 3.3 70B €0.30·€0.50, gpt-oss-120b €0.06·€0.30, DeepSeek V4-Flash €0.15·€0.35, MiniMax M3 €0.25·€1.00, GLM-5.2 €0.45·€1.40, Qwen3.5 397B €0.40·€2.50. Batch tier −50%.
- **EU-Sovereign tier**: routing restricted to attested machines physically in the EU, zero retention, DPA available — e.g. 70B-class €0.35·€0.55 vs IONOS €0.65·€0.65 / OVHcloud €0.67·€0.67 / Scaleway €0.90·€0.90; GLM-5.2 €0.60·€1.80 vs Scaleway €1.80·€5.50 (their public prices, Aug 2026).
- Prompts/completions are never written to disk and never used for training.

## For hardware owners (hosts)

- Install `mahout` (open source; the name means "elephant driver"). NVIDIA from 8GB VRAM and AMD (RX 7900 XT/XTX+, ROCm/Vulkan) supported from day one. Linux first; Windows/macOS follow.
- Machines are benchmarked, not trusted: capabilities are measured, work is verified via activation fingerprints (TOPLOC) and sampled replays; signed receipts are Merkle-anchored publicly every 10 minutes.
- Pay: 80% of token revenue for work actually served; monthly SEPA payout from €50 via Stripe; guaranteed starter work in week one; estimated gross €60–290/mo for well-utilized consumer cards (not guaranteed — depends on demand).
- Safety: sandboxed signed engines only, zero engine network egress, no client code ever, no inbound ports, one-click pause / auto-pause while gaming.

## Sections

- [Homepage](https://elephantpool.ai/)
- [How It Works](https://elephantpool.ai/#how-it-works): request flow + public 80/20 split.
- [Models & Pricing](https://elephantpool.ai/#models)
- [For GPU Owners](https://elephantpool.ai/#owners) and [Earnings Estimator](https://elephantpool.ai/#calculator)
- [Transparency](https://elephantpool.ai/#transparency): verified compute, public receipts, open docs.
- [Whitepaper](https://elephantpool.ai/WHITEPAPER.md) · [Technical spec](https://elephantpool.ai/TECHNICAL-SPEC.md)
- [Full LLM content](https://elephantpool.ai/llms-full.txt)

## Contact

- Email: contact@elephantpool.ai (subject "API early access" or "Host early access")
- Operator: De Buck Technologies, Belgium
- Response time: within 24 hours