Early access. The mesh is live and under active development — latency and uptime vary. Contributors earn credits (not cash yet). Security model →
ClosedMesh
Open peer-to-peer LLM mesh.
Contribute compute

Run a node. Serve the mesh. Earn credits.

ClosedMesh routes chat and API traffic to peers whose hardware fits each model. Contributors who serve real sessions accumulate credits — tracked now, redeemable when the payout rail ships. No crypto token, no signup wall, no promise of instant cash.

Reading the mesh…
Download desktop appInstall CLI runtimeLive status →

How credits work (early access)

The runtime counts completion tokens your machine serves to mesh requests over a rolling 7-day window. The desktop dashboard multiplies that tally by illustrative per-tier rates to show what your contribution will be worth — not money owed today.

Daily driver
$0.05 / million tokens

8B–14B class models at chat-viable latency. Highest volume, most contributors.

Example: 5.0M tokens served → ~$0.25 illustrative

Capacity
$0.25 / million tokens

32B–70B models that fit on beefy solo peers or pooled splits. Scarcer capacity, higher illustrative rate.

Example: 0.5M tokens served → ~$0.13 illustrative

Illustrative only — not a payout. Credits accrue on the public ledger when your peer serves mesh chat traffic. Redemption is manual for the first cohort until automated payouts ship. Early-node multipliers for the first ~100 peers are planned.
Loading credits leaderboard…
Check your credits

Find your node id on /status and paste it to see your accrued early-access credits.

Why run a node?

  • Turn idle hardware into mesh capacity. An M-series Mac with 16GB+ unified memory or a CUDA box with 12GB+ VRAM can serve daily-driver models while you sleep.
  • See contribution in the dashboard. Tokens served, models loaded, uptime, and estimated credits appear in the desktop app's local controller.
  • Open runtime you can audit. Same stack whether you contribute or run entirely for yourself. Read the threat model before sharing a machine you care about.

Hardware that works well

PlatformBackendSweet spot
Apple Silicon MacMetalQwen3-8B solo; 30B+ with enough unified memory
Linux / Windows NVIDIACUDADaily-driver + capacity models on 12GB+ VRAM
AMD / Intel GPUVulkan / ROCmDaily-driver tier; join the mesh, measure on /status

The installer at closedmesh.com/install detects your platform and pulls the matching runtime build.

Get started in three steps

  1. 1
    Install

    Download the desktop app or run the install script. The runtime autostarts and joins the public mesh via an embedded invite token.

  2. 2
    Share a model

    Pick a model your hardware fits — start with Qwen3-8B-Q4_K_M if unsure. The dashboard shows load progress and when you're serving.

  3. 3
    Watch credits accrue

    Serve mesh chat or API traffic. The dashboard's earnings preview updates from real completion-token counts. Check /status to see yourself on the public catalog.

Questions or want to join the first operator cohort? Open a discussion on GitHub Discussions. Ship log: /updates.