See connected devices
Check which computers are online and what resources each one makes available.
The platform
One pool for computers, models, agents, and storage.
PoolParty connects supported computers so you can run models, send jobs to compatible hardware, and use encrypted storage through your apps and agents. Each machine keeps its own measured resources and permissions.
- Connect computers
- Access models
- Share compute
Your Pool · browser workspace in development
Manage your pool from a browser.
The planned workspace brings approved devices, models, agents, storage, compute and conversations into one interface. Sign-in and approval from an installed computer will be required before browser controls can reach your pool.
Set up models and agents
Onboarding for local models, supported provider APIs, agent roles and harness connections is being developed.
Manage pool resources
Review storage, memory and compute per machine, with clear availability and permission states.
The desktop pool remains account-free for local use. Browser access has a separate account and device-approval step. Explore Your Pool and its planned setup ↗
01
Join your computers into one pool
You install the app and join with a code. Machines you own — or your company owns — form a pool. There is no stranger in it, and no sign-up funnel in front of it.
- No passwords, ever. A credential held by the device is the permanent sign-in. There is no login screen to phish and no account to breach.
- One Join button. Paste a code, and you are in. First run offers a one-press recommended setup, with every knob available beside it.
- Joining is not file access. Two tiers of trust: a machine can contribute compute without ever being able to read pool files. Read access requires approval from a device you already trust.
How it works
From separate computers to one private pool
- InstallYour first computer starts a pool.
- Join with a codeAdd the computers you choose.
- Use one networkDiscover their models and resources.
Your computers · one network Owner-controlled
Your laptopAsk from here
Your desktopRun compatible work
Another computerContribute resources
Your PoolParty poolConnect computers. Connect agents. Connect models.
- AI models
- Compute jobs
- Encrypted files
02
Choose when each computer contributes
Contribution is opt-in per machine and gated by conditions that are checked continuously, not once at setup. If a gate closes, work stops — and you are told which gate and why.
Interactive example
See when contribution can run
Change a condition to see what the pool allows. This example does not control your computer. Contribution stopped
Owner permission is off.
Example pool decision Idle
The machine has been unattended for a set period. Come back and the gate closes.
On AC
Plugged in, and above a battery floor. Your laptop is not quietly draining itself.
Thermal
Within thermal limits. This gate is never waived, on any platform, by any preset.
Preemption
Touch the keyboard and running work is stopped immediately — not finished first.
Storage budget
You set a ceiling in bytes. Placement past it is refused. Lower it and data drains off gracefully.
Kill switch
One toggle stops everything. Deleting the policy file is a harder stop still.
What runs by default is sandboxed. The default consent tier executes only sandboxed WebAssembly workloads. A native tier — the one that can use a graphics card directly — is a separate, explicit opt-in, because its sandbox is weaker and the consent says so.
03
Use models, compute, and encrypted storage
The app can use models and other resources available in the pool. Model selection chooses an eligible model and machine; separate computers do not combine their GPU memory into one larger model. Provider onboarding and browser-based resource controls are being developed.
How it works
The job goes where the right GPU is
- Submit a jobExample: CUDA, at least 16 GB VRAM.
- Match hardwareCheck measured capacity and consent.
- Compatible GPURun the payload on that machine.
- Chat with available models served by your own machines, with streaming responses and a visible model-selection reason.
- Run jobs as groups that fan out across eligible machines; every shard is verified server-side before the group is accepted.
- Seal files into the pool and read them back on any device holding the key.
One honest caveat, stated where it applies. Job results are verified structurally — the system checks that a shard did what it was asked to do. That is not the same as checking whether a language model's answer is correct. Model output is owner-trusted, and the app says so on the screen where you read it.
Use your pool
Choose an experience.
Chat, MCP access and device messaging are available in the beta. Agent crew orchestration, provider onboarding and the connected browser workspace are in development.
The privacy architecture
Control where your data goes
These are claims about how the system is built, not promises about how it is operated. The difference matters, so here is the mechanism.
How it works
Encrypt first. Then distribute the pieces.
- Your fileStarts on an approved device.
- Encrypt & splitEncryption happens before upload.
- Pool storageMachines store encrypted chunks.
Run prompts and jobs on your computers
In a personal or company pool, requests travel between machines the owner controls. There is no third-party coordinator sitting in the middle decrypting your prompts in order to route or bill them, because there is no third party in the path at all.
Encrypt files before storage
A file is encrypted on your device before anything leaves it. The chunks that leave are ciphertext. The machines storing them hold opaque blocks — no filenames, no grouping, no keys — and no machine ever holds enough of a file to rebuild even its encrypted form.
Keep file keys off the coordinator
It holds encrypted manifests and placement records. It never holds a decryption key and cannot read pool contents. It does see sizes, counts, and timing — that is a real limit and it is stated in the app as well as here.
Approve which devices can read files
Keys live in the secure key storage of devices you have enrolled, and are moved to a new device only through an approval you perform from a device you already trust. Recovery codes are generated once and never stored by PoolParty.
The trade this design makes. There is no back door and no reset link, because either one would be a way in for someone who is not you. Lose every enrolled device and your recovery codes, and the data is unrecoverable — permanently. PoolParty is not a backup product and does not present itself as one.
Remote access · Coming after beta
Reach your pool across networks
Connect your own computers from another network, and use a phone or laptop as a consume-only device. Both are coming after the beta and are not included yet.
PoolParty
Your pool, from another network.
Cross-network access is planned, with direct connections where possible and a relay you control as a fallback.
- Coming after beta
- Device credentials
- Owner permissions
Set up your own compute pool
Local desktop use is free on hardware you already own and does not require a website account. The planned browser workspace uses separate sign-in and installed-device approval.