The MAI Browser Mining Demo requires a desktop or laptop computer with a dedicated GPU.
Please visit this page on your desktop browser to experience AI mining.
← Back to HomeYour GPU runs real AI inference directly in the browser — no installation, no servers. This is a preview of how the MAI mining network will operate at full scale. A selection of quantized models is available here for testing — the full client will support many more.
Model will be downloaded to your browser cache (~260 MB – 3.4 GB). Subsequent sessions load instantly.
MAI browser mining runs real AI inference on your GPU — exactly how MAI Network nodes operate. Check what you need below.
Mobile browsers don't support WebGPU with full GPU access — browser AI mining isn't possible on phones yet.
The MAI Network native mobile client will use direct GPU APIs for efficient on-device mining.
If your browser doesn't support WebGPU, the demo automatically falls back to WebAssembly (WASM) — CPU-based inference. It's 5–10× slower but still works.
MAI earned is calculated from real compute work your device performs:
This demo runs real AI inference on your GPU — the actual computation is genuine. The MAI earnings display is a simulation to show how the production mining client will calculate rewards.
Your browser uses the WebGPU API to run AI computations directly on your graphics card. The model executes as shader programs — the same math data centers charge thousands of dollars per hour for, now running on your hardware for free.
The AI model (260 MB – 3.4 GB) downloads once into your browser cache. After that, every session starts in seconds — no internet needed. The model runs 100% on your device. Your questions and answers never leave your machine.
Every word the AI generates is a unit of real compute work. In the MAI Network, miners earn MAI tokens for exactly this — processing AI requests for users. Mining Speed (ops/sec) is your hashrate: the faster your GPU, the more you earn.
The power slider and NFT selector show how the production MAI client will calculate your rewards. In the real network, power share (25–100%) controls how much compute you allocate to the network; presale NFTs give a permanent bonus: Bronze +5%, Silver/Airdrop +10%, Gold +15%, Platinum +20%. In this demo these settings update the earnings display only — the actual AI inference runs at full browser capacity regardless.
Zero data is sent to MAI Network servers during inference. The network verifies your work via cryptographic Proof-of-Useful-Work (PoUW) — a hash of your compute output — without ever seeing your prompts or responses.
This demo uses the same AI runtime as the future MAI client. In production: your node connects to the MAI peer network on Solana, processes real user requests, submits PoUW proofs, and receives on-chain MAI rewards weekly.
Yes — the AI model uses VRAM and GPU compute while it is generating a response. When the model is idle (waiting for your next message), it releases resources. You can stop the session anytime with one click. In the production MAI client, power share settings will let you limit how much compute you dedicate to the network.
No — the AI inference running on your GPU is real, but the MAI earnings displayed are a simulation. They demonstrate how rewards will be calculated in the production client. Real MAI tokens will be paid on-chain after the mainnet launch. Presale participants get early access to the mining client.
SmolLM2 1.7B — best for most users, stable and fast. DeepSeek R1 1.5B — shows live reasoning steps, great for complex questions. Qwen 2.5 1.5B — best multilingual support. SmolLM2 360M — fastest, works on older or integrated GPUs.
No. The model is cached in your browser after the first download. Subsequent sessions load in seconds. It only re-downloads if you clear browser data or manually delete the cache from the interface using the trash icon.
In the production network, miners earn MAI tokens for processing real AI inference requests from other users. Your reward depends on the volume of useful work your node completes — measured in compute operations (PoUW). NFT holders receive a permanent percentage bonus on top of base earnings.
MAI Network is infrastructure, not a single AI. The network routes requests to the best available open-source model running on miners' hardware worldwide. This makes it censorship-resistant, upgradeable, and decentralized — no single model or company controls it. You can see exactly which model you are talking to.
DEMO VERSION — FOR DEMONSTRATION PURPOSES ONLY. All AI computations run entirely on your device using your CPU/GPU and local memory — no data is transmitted to external servers. The AI responses are generated by open-source language models and may be inaccurate, incomplete, or inappropriate. The MAI earnings displayed are a simulation only — they demonstrate how the production MAI Network client will calculate rewards and do not represent actual token accruals, financial returns, or any binding commitment by MAI Network. MAI Network provides this demo solely to illustrate the concept of decentralized local AI inference. AI models used in this demo are open-source and subject to their respective licenses (Apache 2.0, MIT, Gemma Terms of Use) — compliance with model licenses is the responsibility of the user. MAI Network is not responsible for any actions taken based on AI-generated content, any direct or indirect damages arising from the use of this demo, any loss of data, system instability, or performance degradation caused by running AI models locally, or any misuse of this software. By using this demo you confirm you are at least 18 years of age and acknowledge that you do so entirely at your own risk. This software is provided "as is" without warranty of any kind.
