Connecting…

LIMONATA INTELLIGENCE · TESTNET

Intelligence powered
by people.

Ask a question, put your GPU to useful work, and explore contributor activity. Contributors keep portions of Qwen3-32B on their GPUs. The shared chat becomes available when every required portion is ready.

Open chat Contribute computeExplore contributors
Ready GPU contributions—Loaded portions and independent models
Available models—Selection follows availability
Active chats—Shared and independent requests
Provider rewardsFunded work onlyPaid requests credit accepted work

CHAT

Ask the network

Model and payment are shown before a request
Limonata AIFinding providers…

Start with a question.

Choose a ready model and see which one answered. Responses may be inaccurate and do not include live blockchain data.

Enter to send · Shift + Enter for a new line0 / 1800

CONTRIBUTE / 01

Your GPU can do
useful work.

Connect a compatible GPU and keep your assigned portion of a shared model in memory. Your provider panel shows its state and the accepted requests your GPU helped complete.

Checking contribution capacity…

  1. ConnectStart the client with a public payout address. No wallet private key is needed.
  2. PrepareDownload your assigned portion once. It stays loaded while the client runs; reconnecting does not reload it.
  3. ComputeYour GPU executes its part of the model. Extra contributors prepare spare copies without interrupting active chats.
  4. CollectShared requests are currently sponsored and earn no LIMO. Paid independent requests retain their existing rewards.

START CONTRIBUTING

Windows provides one-click setup. The Linux client checks the NVIDIA GPU before joining. The client detects the device name and memory automatically.

Windows client ↗Linux / Runpod ↗

Windows 0.8.0 · AMD / NVIDIA Vulkan. Linux 0.8.0 · NVIDIA CUDA, Ubuntu 22.04+ / x86-64.

A measured AI work rate.

Explore accepted requests and historical response delivery by model. The explorer measures characters per second from accepted responses; GPU memory tells us which models fit. Hardware capacity alone does not determine speed. Individual token rates are identified as provider reported.

Who funds rewards?

For a paid request, the user deposits LIMO in escrow. Accepted work credits the provider. Sponsored requests are free and currently pay no provider reward. Shared-model requests in this release are sponsored; distribution of paid rewards across their GPUs is not enabled.

CONTRIBUTOR EXPLORER

Follow the work behind the network.

See contributors by public payout address, their accepted work and credited rewards. Pending settlement is separate; the withdrawable escrow balance is shown only when verified.

View all contributors ↗

Testnet uses a trusted coordinator to accept results. Test LIMO has no monetary value. No public inbound GPU port is required.

0.8.0 release notesWindows guideLinux guideWindows sourceLinux sourceWindows SHA-256Linux SHA-256