A crew of AI bots that lives on your iPhone and works from its own cloud computer. Powered by open-weight models.
A working preview of the iOS app. Tap around. Bot replies are scripted for the demo.
One app and one tap to start your cloud computer. No API keys, no terminal, no Mac left running overnight.
Install Beans Bot on your iPhone and sign in with Apple. That's the whole account setup.
Your own cloud computer spins up with a browser, files and a scheduler. It stays on, so your bots do too.
Give it a name and a job. Leave the brain on Auto, or pick an open-weight model yourself.
Start with one clear request and check the answer. Anything that sends or pays waits for your OK.
Give each bot a job, a brain, and a computer. They work in parallel in the cloud and come back to your phone when something needs you.
A private cloud VM with a real browser, terminal and files. Bots work there side by side, and you can watch the screen live from your phone.
Sending, paying, posting and deleting always wait for you. Approve from the notification without opening the app.
Each bot can use a different open-weight model. Use a big one for planning and a small fast one for quick replies. Bots keep their memory when you switch.
The app sends tasks and shows results. Nothing heavy runs on your iPhone, so your battery won't notice.
Every account gets its own isolated cloud VM. Close the app or lose signal, and the work keeps going.
Each step goes to an open-weight model on a fast inference provider. No GPU, no downloads, no lock-in.
Pick one to see what it reports back, and what it waits for you to approve.
Example conversation. Names are made up.
There's no model on your phone and you don't need one on your desk. Beans Bot sends each step to an open-weight model that's good at it.
| Model | Made by | Good at | Default for |
|---|---|---|---|
| Kimi K2 | Moonshot AI | Long agent runs, tool use | planning |
| DeepSeek V3.2 | DeepSeek | Research, long documents | research |
| Qwen3 Coder | Alibaba Qwen | Repos, scripts, fixes | code |
| GLM-4.6 | Z.ai | Balanced everyday work | general |
| gpt-oss-20b | OpenAI | Fast, cheap replies | quick replies |
Served by hosted inference providers. You can also bring your own provider key, which is stored write-only on your VM.
No. Nothing runs a model on your phone or computer. Beans Bot uses open-weight models served by hosted inference providers, so you get large-model quality without owning a GPU.
Bots are only useful if they can actually do things: open a browser, keep files, run on a schedule. A dedicated VM gives your crew a real computer that stays on, isolated from every other account.
Read, research, draft and organize. Anything that sends, pays, posts or deletes waits for your approval, in the app or from the notification.
On your VM. Logins and keys go into a write-only vault there. Prompts are sent to the inference provider that runs the model for that step.
Yes. Use hosted inference out of the box, or add your own provider key in Settings.
iOS comes first because that's where approvals and notifications work best. Other platforms are on the roadmap.
The beta is free for early users. We'll publish plans well before the beta ends.
We're letting people into the iOS beta in small batches.