stats
Live stats
Utilization, VRAM, temperature, power and fan for each GPU, plus CPU, RAM and disk. Updated every second.
gpu 0 87% · 18.2/24 GB · 71 °C
Your GPU rig,
in your pocket.
Live stats, job alerts and one-tap actions for the machine with the GPU. On your phone, from anywhere, with nothing to set up.
- encryption
- End-to-end, phone to machine. QUIC + TLS 1.3
- our servers
- See nothing. Relays pass on bytes they can't read
- network
- No open ports, no VPN. Works behind NAT and CGNAT
- runs as
- Your user, never root
- price
- Free for one machine
- rig-1
- online · direct 18 ms
- link
- end-to-end encrypted
- gpu 0
- 87%
- vram
- 18.2 / 24 GB
- temp
- 71 °C
- power
- 380 W
- job 12
- train.py · epoch 31/40
- eta
- 38 min
what it does
Everything you'd walk to your desk for.
alerts
Job alerts
Wrap any command in pgpu run. Your phone buzzes when it finishes or crashes, with the exit code and run time.
job 12 exit 0 · 2h 14m
actions
Quick actions
Save the commands you run all the time. One tap from your phone, Face ID for the risky ones.
restart ollama ok in 1.2 s
processes
GPU processes
See which process holds which GPU and how much VRAM it uses. Stop a runaway one from your phone.
pid 48213 python · 14.1 GB
terminal in 1.1
Terminal
A real shell on the machine, as your user. Check a log, restart a service, answer Claude Code.
gan@rig-1:~$ tail -f train.log
models in 1.1
Local models
Chat with the Ollama models on your own GPU. Pull, load and delete them from your phone.
qwen3:14b 42 tok/s
ai agents
Works with Claude Code and Codex.
Run a coding agent, or any long task, through pgpu run. Your phone gets the result when it ends, so you don't have to sit and watch it.
From 1.1, open the terminal on your phone to answer an agent that is waiting for you.
gan@rig-1 ~/appbash
PocketGPUnow
Job finished · rig-1
fix tests: claude exited 0 after 18m.
setup
Installed and paired in a minute.
1 install
$ curl -fsSL https://pocketgpu.app/install | sh
Installs to ~/.local/bin and starts a user service. No sudo.
2 pair
$ pgpu pair
Shows a QR code. Scan it in the app and choose view, control or shell.
3 watch
$ pgpu run -- python train.py
Your phone tracks the job and tells you when it ends.
or docker
$ docker run -d --gpus all -v pocketgpu:/data ghcr.io/gato25/pocketgpu
For RunPod, Vast.ai and other GPU hosts. No ports needed.
linux x86_64 or arm64 · wsl2 and docker · nvidia driver for gpu stats · native windows in 1.2
security
End-to-end encrypted.
We can't see your rig.
Your phone and your machine each hold a key. Every connection is encrypted with QUIC and TLS 1.3 from one end to the other: direct when it can be, through a relay when it can't. Relays only pass along bytes they can't decrypt.
Your phone paired key a91c…
direct · quic + tls 1.3
rig-1 agent · your user
or through a relay that sees encrypted bytes only
datagoeswho can read it
Stats, logs and job output Your phone, end-to-end encrypted readable byonly you
Terminal and chats (1.1) Your phone, end-to-end encrypted readable byonly you
Alert text, like “Job finished” Push server, then Apple or Google readable bydelivered, never stored
Push token Push server readable byus, to deliver alerts
Waitlist email Cloudflare and our email provider readable byus
- paired keys only
- The agent drops unknown keys before reading a single message.
- permissions
- View, control or shell, checked on every request.
- never root
- The agent refuses to run as root.
- face id
- Asked before kills, risky actions and every terminal session.
- audit log
- Every action and refusal, kept 90 days on your machine.
- open source
- Read every line that runs on your machine. MIT or Apache-2.0.
pricing
Free for one machine.
Pro adds machines, a month of history and every alert type, for $4 a month or $36 a year. Every plan includes the open-source agent and end-to-end encryption.
| Feature | free | pro | team · later |
|---|---|---|---|
| Price | $0 forever | $4 a month, or $36 a year | $12 per seat a month |
| Machines | 1 | unlimited | shared, 1 to 5 people |
| History | 24 hours | 30 days | 30 days |
| Stats, jobs, quick actions, processes | Included | Included | Included |
| Terminal and local models (1.1) | Included | Included | Included |
| Job finished and crashed alerts | Included | Included | Included |
| Temperature, disk, offline and Ollama alerts | Not included | Included | Included |
| Custom alert thresholds | Not included | Included | Included |
| Home-screen widget and Live Activity (1.2) | Not included | Included | Included |
| Roles, audit export, alerts to the whole team | Not included | Not included | Included |
| Custom relays and webhook alerts | Included | Included | Included |
| Join the waitlist | Join the waitlist | Get notified |
questions
Common questions.
Do I need to open ports or set up a VPN?
No. The phone and the agent connect directly when they can and through an encrypted relay when they can't. There is nothing to configure.
Can you see my data?
No. Stats, logs, the terminal and chats are end-to-end encrypted between your phone and your machine. Alerts carry only a short text like “Job finished · rig-1”.
Which machines work?
Linux with an NVIDIA GPU, including WSL2, Docker, RunPod and Vast.ai. Without a GPU you still get CPU, RAM and disk. Native Windows comes in 1.2.
Does it work with Claude Code and Codex?
Yes. Run them with pgpu run to get an alert when they finish. The terminal in 1.1 lets you answer them from your phone.
Is the agent open source?
Yes, under MIT or Apache-2.0. You can read every line that runs on your machine before you install it.
When can I use it?
The private beta for iOS and Android opens to the waitlist first. Join, and we'll email you when it's ready.
beta
Get the beta on your phone.
iOS and Android. Confirm your email and we'll write when the beta opens. Every email has a one-click unsubscribe.