Decentralized AI · No data center

Your phone works for the world while you sleep.

Meridian Moonlight turns idle phones into a global AI network. Contribute only while you're charging, on Wi-Fi, and asleep — and everyone gets free AI in return. No data center. No token. No cost to you.

Pledge a phone See how it works
Follow-the-moon · global device availability Simulated 24-hour cycle
Time (UTC)00:00
Fleet available—
Night-side regions—
1.4%Adoption needed

That's when this beats the largest data center on Earth.

Around 2.2 billion phones can already run a modern small AI model. If just 1.4% of them join — about 30 million devices — the network's sustained compute passes the biggest AI data center ever built. At full participation it's roughly seventy times larger.

How it works

Whole models on each phone. Night does the scheduling.

On your device

A complete model, not a fragment

Every phone runs an entire small AI model locally. Nothing is split across the internet, so there's no lag from shuttling data between devices — and your prompts never leave your hardware.

Only when idle

Charging, on Wi-Fi, screen off

Contribution is capped to conditions where you'd never notice it. No battery drain, no mobile data, no heat while you're using your phone. Switch it off any time.

Around the clock

Somewhere it's always 3 a.m.

Demand peaks during the working day; supply peaks overnight. Since night circles the planet continuously, the network is always fed by the hemisphere that's asleep.

What it's for

Free AI for people who can't buy it. And a research instrument for everyone else.

Every day

An assistant that costs nothing

A tutor, translator, and general assistant for the billions of people locked out of paid AI by price or geography — running privately on the device in their hand, with the network as backup.

Every night

Open science while the world sleeps

Idle hours become batch compute for public research: drug candidate screening, protein dynamics, materials discovery, climate ensembles. One night at full scale exceeds a year of Folding@home at its peak.

Honest limits

What this will never do.

Every project in this space leads with what it can do. Here's what physics forbids, stated up front — because a plan you can't poke holes in isn't a plan.

  • It is not one giant brain. Splitting a single large model across phones on the internet fails — generating each token means passing data through every layer in sequence, and network hops are millions of times slower than the wiring inside a chip. So we don't split models at all.
  • It will not train frontier models from scratch. Training needs terabytes per second of bandwidth between processors sitting inches apart. No amount of adoption changes that.
  • The compute is throughput, not one pooled engine. Millions of independent tasks at once — not a single task made a million times faster. That's the right shape for serving people and running science, and the wrong shape for anything else.

The promise

No token. No wallet. No ads. Ever.

Most decentralized compute networks pay contributors in cryptocurrency and sell the compute back to developers. Moonlight isn't a marketplace. It's a public utility: free to use, free to join, and open source end to end. Whatever eventually pays the hosting bill, it will never be your access, your attention, or your data.

Roadmap

Every phase ships something that runs.

M0

One node lives Building now

A single phone runs a 3B model on its charger and reports its available compute to a coordinator. Proof the atomic unit works.

1 device · Month 0–1
M1

The network answers

Requests route between volunteers' devices, with duplicate execution to verify results and a live map of who's online.

~100 devices · Month 1–3
M2

Follow the moon

Region-aware routing, a published 24-hour availability curve measured from real nodes, and the first overnight science batch job.

~10K devices · Month 3–9
M3

Open protocol

Specification v1.0 so anyone can build a compatible node, federated model improvement, and a governance model that outlives its founder.

~100K–1M · Month 9–18
M4

Public utility

Full peer-to-peer discovery and an institutional research program. At roughly 30 million devices, the network passes the largest data center on Earth.

1M → 30M+ · Year 2–3

Join

Pledge a phone.

No app to install yet. Leave an email and you'll hear once — when the first node is ready to test.

Builders

Read the code →

Android, Kotlin, llama.cpp, and a Node coordinator. Milestone 0 is a weekend project.

 Skeptics

Check the math →

Every compute figure on this page is derived in the open. Find an error and file it.

 Researchers

Bring a workload →

If you have an embarrassingly parallel problem and no budget, that's exactly the point.

Keeping it alive

There's no donate button yet — on purpose.

Asking for money before there's a working node felt like the wrong order. But the project does have costs: developer accounts, a small server, and test phones that match what most contributors actually carry.

If you'd want to help cover those, say so in your email when you pledge a phone above, or write to hello@meridianmoonlight.com. Nothing is set up to accept money right now — this is just a headcount, so I know whether to bother building that path.