Limonata ProviderTESTNET PREVIEW

YOUR HARDWARE. SHARED COMPUTE.

Put your GPU
to work on AI.

Join the shared AI model with one click. Your GPU keeps an assigned part of Qwen3 32B ready and processes real requests with other contributors.

START CONTRIBUTING

Your GPU becomes part of one shared model.

Setup, stage assignment and a real GPU check run for you. The complete model becomes available when every required stage is ready. Shared-model work is currently sponsored and earns no LIMO.

Your identity is saved on this Windows account. No wallet private key is needed.

  1. Verify files
  2. Check GPU
  3. Connect
  4. AI work
Connecting…

Opening your private local session.

0 completed 0 failed

YOUR COMPUTE

Real work. Measured contribution.

Offline
Historical response rate0 tok/s

Waiting for your first request.

Completed output tokens per second, including prompt processing. Live speed is not measured.

Escrow balance available to withdraw—

Test LIMO · includes credits and refunds

Shared by machines using this payout address.

Earned total—

Test LIMO · confirmed in escrow

Recorded settled earnings, including previous withdrawals. Separate from your escrow balance.

Awaiting settlement—

Test LIMO · accepted paid requests

Only confirmed settlements enter your credited total.

Accepted requests—
Output tokens supported—

Average speed appears after measured work.

Reward per paid request—

Connect to check the current rate.

Reward totals come from the coordinator. Sponsored requests earn no provider LIMO.

Use the wallet matching your payout address. Your wallet will ask you to confirm the transaction and network fee.

Advanced options

WORK SETTINGS

Contribution preferences

Shared mode keeps your assigned layers resident between requests and reconnects. Additional compatible GPUs provide stage capacity or warm replicas. Independent mode serves a complete smaller model on one GPU. Legacy tensor RPC remains available explicitly.

GPU inference will be checked when you start.

Shared stages require sufficient free VRAM for their assigned layers and conversations. A GPU can wait for compatible work without contributing to current model readiness. CPU fallback is rejected.

FILES AND CACHE

Verified local files

Shared mode downloads only the assigned stage: about 14.99 GB or 5.86 GB, plus verified runtime files. Allow at least 18 GB or 8 GB of free disk space. Interrupted transfers resume; complete files are verified before loading and reused on later starts. Independent mode separately chooses a complete 1.5B, 8B or 32B model.

This limit applies only to legacy RPC transfers. Shared stages and independent whole models have separate verified disk files. Chats are never cached here.

Windows 10/11 x64. Some systems also need the Microsoft Visual C++ 2015–2022 x64 runtime. GPU drivers are never installed automatically.