MESH INFERENCE · TEAM SE7EN

Your phone can't run
a big model.
Your room can.

Phones, laptops, the PC in the corner: one code away from becoming a single machine. Each holds a slice of the model; together they run the whole thing. Nothing to install.

nothing to install· no account· nothing you type reaches a server· keeps going when a device walks away
WHAT IS AI SWARM

One model,
shared by the room.

A model needs more memory than one everyday device has. AI Swarm lets a room of them share the load — and, unlike dealing layers by whoever pledged most, it measures each device and solves the placement.

room CCPU
01

Open a link

A four-letter code, or scan the QR on someone's screen. No install, no sign-up, no drivers — a locked-down work laptop can join.

02

Chip in what you like

You choose how much of your device to lend. It gets measured, not trusted: real speed, real free memory, real latency to every other device.

03

Think together

Every word laps the room and appears on every screen at once. Take your device back whenever — the room re-plans and carries on.

Private by design. The signalling server introduces devices and then stops mattering: it has no message type that could carry a prompt.
5.44 tok/sQwen3 0.6B split across two devices, weights streamed from a CDNrecorded run · README status table
49.5 msmedian network lap for one word to cross the room and come backsame run
13.1 sto heal after a device left 58 tokens into an answer — and finish the same sentencerecorded recovery run
295tests, covering the wire format, the solver, recovery and deliverynpm test
Your devices are already a cluster.
Introduce them.
start a swarm