Got a Mac Mini or Mac Studio sitting idle after the M4 craze? That 16–512GB of unified memory can run AI models and earn you xRAM tokens. Install the app, share your spare compute, get paid.
Pick a model, fund a session, and keep aggregate Mac RAM focused on your workload instead of waiting through cold starts every time.
Install the app, choose how much memory to share, and help keep larger models online across distributed Apple Silicon nodes.
These numbers come from the same live onboarding endpoint the user and provider funnels consume.
Loading live network status...
Whether you want to use AI or power it, RAM Aggregator has you covered.
Link MetaMask and deposit xRAM tokens — or try for free with the demo key.
Pick from Llama, Mistral, Qwen, DeepSeek and more — from 1B to 671B params. See live pricing and provider availability.
Your prompts are routed to the best available provider. Pay per token, way cheaper than centralized APIs.
Run curl -fsSL ramaggregator.com/install | bash in Terminal. More options →
Choose which LLMs to serve. Models auto-download from HuggingFace and run natively via MLX. Use the RAM slider to cap resource usage.
Every inference you complete earns xRAM tokens — minted on-chain to your wallet. Your Mac is literally printing money.
Enterprise-grade infrastructure, decentralized by design.
Drop-in replacement for any OpenAI SDK. Point your base_url at RAM Aggregator and your existing code just works. Native chat templates for every model family.
Models run via MLX — Apple's ML framework optimized for M-series chips. Blazing fast 4-bit quantized inference with threaded execution for reliable connections.
Models too large for one Mac are automatically split across multiple providers. A 405B-parameter model can run across three 96 GB machines. Completely transparent to users.
Optional E2E encryption (X25519 + AES-256-GCM) ensures the coordinator never sees your prompts. Only you and the provider can read your data.
xRAM is a real ERC-20 token on Base L2. Emission rewards are minted directly to your wallet via smart contract. Bitcoin-style halving — early providers earn the most.
RAM allocation slider, Prevent Sleep toggle, and model management. Control exactly how much of your Mac to share. Auto-updates keep your node current.
Prices adjust based on real-time network utilization. Low demand = cheaper inference. High demand = more provider rewards. Floor and ceiling protect both sides.
Llama 3.x, Qwen3, Mistral, DeepSeek R1, and more. MoE models run at surprising speeds. Smart routing picks the best provider with the model already loaded.
No GPU datacenter. No single point of failure. The network is powered by individual Mac owners worldwide sharing their idle compute.
We're upfront about what's encrypted, what's visible, and to whom. No fine print.
When E2E encryption is enabled, your prompts and AI responses are encrypted in your browser before leaving your device. The coordinator only sees opaque ciphertext.
No email, no name, no cookies, no analytics. You connect with a wallet address — that's it.
Your prompts are never used to train, fine-tune, or improve any AI models. Ever.
Every connection is encrypted in transit with modern TLS. E2E adds a second layer on top.
The provider running your inference must see the plaintext to generate a response. This is how all AI inference works — including centralized services like ChatGPT and Claude.
The coordinator sees which model you used, payload sizes, timing, and your wallet address — but not your prompt or response content (with E2E on).
Prompts exist only in memory during inference and are discarded immediately. No logging, no history, no retention.
Your wallet address is not shared with providers. With E2E, each request uses a fresh ephemeral key — requests can't be linked to the same user.
A deflationary supply with halving epochs, designed to reward early providers.
Fixed supply. No inflation. Every xRAM token is either in the treasury waiting to be earned, or already in a provider's wallet.
Rewards halve every 17.5M tokens minted — just like Bitcoin. Early providers earn significantly more.
| Epoch | Reward/Job | Treasury Remaining |
|---|---|---|
| Epoch 1 | Base rate | 70M → 52.5M |
| Epoch 2 | 50% of base | 52.5M → 35M |
| Epoch 3 | 25% of base | 35M → 17.5M |
| Epoch 4 | 12.5% of base | 17.5M → 0 |
Deposit xRAM via MetaMask to an escrow smart contract. Tokens are deducted per inference based on model size and token count. Withdraw your unused balance at any time.
Earn xRAM for every inference job you complete. Rewards are calculated from the emission schedule and minted on-chain. Stake xRAM to increase your earning multiplier and priority in job routing.
Whether you want cheap AI inference or want to earn by sharing your Mac's power, the network is live now.