PlanetCompute turns spare computing power from everyday computers around the world into one shared AI cloud. Instead of relying only on massive data centers, we put the unused capacity of laptops and desktops to work, whether they are sitting idle or quietly helping while their owners carry on with their day.
Every AI job is split into WorkPacks: small, signed pieces of work sized so each one finishes in seconds to minutes on an ordinary computer. Thousands of machines work on one job at once, and their answers are checked and stitched back together.
Open this page and press start, or install the desktop app. You choose how many cores, how much CPU and which hours.
Your machine is sent signed WebAssembly WorkPacks matched to its power: a fast GPU gets more, a laptop gets less.
Work runs only when your machine has room for it, under the CPU cap you set. Use your computer and it backs off.
Results are checked against other machines and random audits. Correct work earns credit; wrong answers earn nothing.
A WorkPack that is too small wastes time on the network; one that is too large leaves everyone waiting for the slowest machine. PlanetCompute sizes each pack by time, for each device, so every pack finishes together and the final answer arrives fast.
PlanetCompute focuses on inference: running trained AI models to make something new. Each workload is cut along the lines where work is naturally independent.
Each picture is generated whole on one machine; many machines make many pictures at once.
one image per unitA shared sketch of tempo, key and melody lets segments of a song be rendered side by side and blended.
10–30 s segmentsKeyframes anchor short chunks of motion, so a whole clip is rendered by many machines in parallel.
keyframe-anchored chunksBatch work like summaries, translation and extraction runs on warm machines that already hold the model.
requests in batchesSeveral candidate solutions are written at once, and computers without a GPU help by running the tests.
candidates + test runsThe same WorkPacks run in both places, so they share one network, one checker and one reward ledger.
Nothing to install. Press start and this tab becomes a volunteer.
A lightweight app built in Rust that works overnight while your computer is idle.
Hidden background computing earned a bad name years ago because it ran without asking. PlanetCompute is the opposite: opt-in, visible, and sealed off from everything on your machine.
Every WorkPack is WebAssembly with no access to your files, camera, microphone or network. It can only do arithmetic.
Each WorkPack is identified by its fingerprint and signed. A tampered pack is rejected before it runs.
Nothing starts without your consent. A visible indicator shows when work is running, and one click pauses it.
Volunteers only ever receive non-confidential inputs. Nothing about you is shipped to other machines.
Each WorkPack goes to several machines on different networks. Once three matching answers from three different networks arrive, the result is accepted.
A fake answer always arrives first, because it costs nothing to compute. That is why answers must match across different networks, and why a fake only ever earns a strike.
Run image, music, video, chat and coding inference on a global pool of everyday machines. Results are checked before you receive them, and finished parts stream back as they arrive.
One job fans out across the compute grid, so wide jobs finish in a fraction of the time.
Matching answers, random audits and penalties keep the output honest.
Uses hardware that already exists instead of building more of it.
Capacity from every time zone: someone's night is always somewhere.
By putting existing hardware to work, PlanetCompute makes AI more accessible, sustainable and global. Hundreds of millions of computers sit idle every night. Together they are one of the largest computers on Earth.
Lower-cost compute opens AI to students, researchers and small teams, not just the largest companies.
Spare capacity is used where it already sits, on machines that are already built and already switched on.
Volunteers who help power the network are credited for every verified piece of work they complete.
Yes. Every WorkPack is WebAssembly, the same sandbox your browser uses for web apps. Baseline packs have no imports at all: they cannot read your files, open network connections or see anything else on your computer. Packs are signed and checked before they run.
You choose a CPU cap (20% by default) and how many cores to share. The desktop app goes further: it works only when you are away, on mains power and inside the hours you allow, and steps aside the moment you come back.
The browser console on this page runs a real demo WorkPack that estimates π: a 133-byte WebAssembly module on your processor's cores, and the same calculation as a WebGPU compute shader on your graphics card, each under the cap you choose. It shows exactly how volunteering works. Demo points are for display only while the live compute grid is being built.
The browser can use your graphics card through WebGPU, but browsers slow down background tabs and WebGPU cannot yet reach every part of a graphics card, such as the tensor cores used for AI. The desktop app is a lightweight Rust program that runs the same WorkPacks overnight, with access to your GPU for AI work.
Spend them on the PlanetCompute compute grid. Credits are exchanged one for one for free processing time, measured in the same units as the work you contributed, so a night of work on your laptop buys a job that can finish in minutes on the grid's graphics cards.
Rewards follow verified work, not time spent. Each WorkPack has a fixed value; the first correct finisher earns the large share and every other correct finisher earns a consolation share. Wrong answers earn nothing.
Mostly electricity. A laptop running all night typically uses a few cents of power; a desktop graphics card more. Running only on mains power and inside your chosen hours keeps you in control of the cost.
Not yet. Phones restrict long background work, so PlanetCompute focuses on laptops, desktops and servers.
Join from your browser in seconds, or download the desktop app and contribute while you sleep.
Both options run the same sandboxed WorkPacks. You can switch any time.
Demo mode: each WorkPack estimates π by Monte Carlo sampling. On the processor it is a 133-byte WebAssembly module in Web Workers with no imports, so it can only do arithmetic; on the graphics card it is the same calculation as a WebGPU compute shader, which the browser validates before it runs. Points are for display only until the live compute grid opens.