Your personal AI cloud — built from the computers you already own
Your devices. Your AI.
Your cloud.
Corewire turns the machines you own into one private AI cloud. Models run on your hardware, answers stream peer-to-peer over an encrypted network only your devices can join — from home, from the office, from a café.
P2P encrypted7 modalitiesFree for 10 devices
01 · Setup
One install. They find each other.
Install on your machines
One daemon on your gaming rig, workstation, or home server. It probes what the hardware can comfortably serve and joins your account's encrypted network.
They find each other
Every device on your account discovers every other, wherever it is. A tiny coordinator introduces them; after that, traffic flows directly, device to device, end-to-end encrypted.
Use your AI from anywhere
Chat with the model on your desktop from your phone on the train. Generate an image on your GPU from a café. Your hardware does the work — nothing runs in someone else's cloud.
02 · Features
What the fleet does
Add a device with one install
Install Corewire on a machine and it joins your personal cloud: no port forwarding, no configuration. It finds the rest of your fleet by itself.
Install models once, use them anywhere
Serve a model on the machine that can hold it, then use it from any device signed into your space: desktop, laptop, or mobile.
Every device in one chat
Through the built-in MCP server, any LLM can call on the whole fleet: ask a chat for an image, a video, or a 3D model, and the right machine serves it.
Roles for people, roles for devices
Members carry roles and permissions; owners control enrolment and removal. Devices declare a role too: client, server, or both.
One open API for the whole fleet
Every node speaks the same standard, open API. Anything that can talk to one machine can talk to all of them, through one address.
A catalog graded to your hardware
A curated catalog of current open models, each graded against the machines you own: the app shows what fits, what it would displace, and serves the best your hardware can hold. New models arrive as catalog data, not as app updates.
03 · Capabilities
Seven modalities, one fleet
Everything below is served by the shipping app today, graded to what each of your machines can hold.
Chat
Local language models, streamed token by token, with tokens per second in the window.
Images
Generation presets from fast to best, graded to what your GPU can hold.
Video
Short clips rendered on your own card, every frame yours.
Voice
Text to speech that runs on any CPU — the modality every machine can serve.
Transcription
Speech to text on your hardware, measured in seconds of audio, kept at home.
Music
Songs composed on your GPU.
3D
Meshes with PBR textures, generated from a prompt.
What ships next
The catalogue is data, not code — as models compress onto smaller hardware, your fleet picks them up without waiting for a release.
04 · Why trust it
Private by architecture, not by promise
Inference never touches our servers. A coordinator introduces your devices and steps out of the way; conversations, images and models stay on hardware you can point at. When two devices cannot reach each other directly, the fallback relay carries only what it cannot read — the traffic stays end-to-end encrypted.
A record you hold, not a bill you trust
Every invocation is recorded on your side — which machine served it, what it was, in counts, bytes and seconds, never estimated money. Export it as CSV or JSON, purge it when you like. It is the receipt for the cloud subscription you stopped paying.
Your agents work your fleet
The built-in MCP server hands your AI tools to any client that speaks it, and the app's own assistant manages the fleet — start a model, pick a GPU, name what will not fit and what would make room.
05 · Pricing
Free for personal use
Personal use is entirely free: a fleet of 10 devices, every modality included, no card, nothing that expires. Pro raises the ceilings; only relayed traffic is ever metered.
See pricing06 · Questions
Answers before you ask
What hardware do I need?
Whatever you have. The app grades each machine and offers what it can comfortably serve — a laptop CPU speaks and chats, a gaming GPU generates images and video. Nothing needs a server rack, and voice runs on anything.
Do I need to open ports or configure my router?
No. Devices find each other through the coordinator and connect directly wherever the network allows. Where it does not — hotel wifi, an office firewall — an encrypted relay carries the difference, and the app tells you when it does.
What leaves my machines?
Coordination metadata: which devices are yours and how to reach them. Prompts, outputs, models and the record of what ran never reach us — there is nothing to harvest and nothing to subpoena.
What does it cost?
Personal use is entirely free: every modality, no trial clock. Pro raises the ceilings for bigger fleets, and only relayed traffic is ever metered — most connections never touch it.
Which platforms?
The daemon runs on Linux, Windows and macOS. The desktop window ships for all three too — the download page states the one glibc floor Linux has, and that the daemon runs below it anyway.
Point it at your hardware
Install on the machine with the GPU first. Everything else joins by signing in.