Nexus Client
Lend your GPU. Earn GNN.
Run an eligible open model on hardware you already own, serve inference jobs for the Nexus network, and get paid in GNN for the compute you contribute.
How it works
- 1
Install & pull a model
Download the Nexus client and pull an eligible model with Ollama. The client detects your GPU and the model tiers it can serve.
- 2
Init & stake
Run
nexus-client initto generate a node key, link your Solana wallet, and stake GNN so the network can trust your node. - 3
Start earning
Start the client and it joins the per-model queue. Completed jobs build reputation and pay out GNN every epoch.
Eligible models
These are the model classes accepted at launch. VRAM figures are honest minimums for usable throughput — more headroom means more concurrent jobs and higher tiers.
| Model | Tiers | Min VRAM | Quant notes |
|---|---|---|---|
| Hermes 3 | 8B | ~10 GB (Q8) / ~6 GB (Q4) | Q8 for best quality on a single consumer card; Q4 fits 8 GB GPUs with a small quality trade-off. |
| Kimi K2 | Large MoE | 2×80 GB-class, or hosted | Sparse mixture-of-experts — not a single-consumer-GPU model. Realistically needs datacenter-class GPUs (e.g. dual 80 GB) or a hosted backend. |
| Qwen | 2.5 — 7B / 14B / 32B | ~8 GB (7B) / ~16 GB (14B) / ~24 GB (32B) | VRAM scales with tier; figures assume Q4–Q5. 7B runs on most modern gaming GPUs; 32B wants a 24 GB+ card. |
Download
Grab the client for your platform. Releases are published on GitHub.
All releases & changelog → · View source on GitHub → · Checksums
Quickstart
Four commands from zero to earning. Replace the model with whichever eligible class your GPU can serve.
# 1. Pull an eligible model
ollama pull hermes3:8b
# 2. Initialize your node (creates ~/.nexus-client/node_key, links wallet)
nexus-client init
# 3. Stake GNN so the network trusts your node
nexus-client stake --amount 1000
# 4. Start serving jobs and earning
nexus-client start Network stats
Live snapshot of the Nexus compute network.