Your GPU first. The cloud when you need it.
Turn your computer into your own AI cloud. Install one app. It checks your hardware, downloads compatible models, and gives you an API your agents and apps can use.
Routine tasks run on your machine. Jobs that need more power can move to cloud GPUs, with an upfront quote paid in SOL.
This machine · GPU · 12 GB · Coding model · 7B loaded
- 09:41:15coding-agentautocompleteLocal47 tok/s
- 09:41:13coding-agentrename symbolLocal52 tok/s
- 09:41:11coding-agentwrite testLocal44 tok/s
- 09:41:09coding-agentexplain errorLocal38 tok/s
- 09:41:07coding-agentcomplete functionLocal41 tok/s
01 / Setup
One install prepares your machine.
$ granio setup
Checks your hardware.
Granio reads your GPU and memory before it downloads anything.
Picks models that fit.
Larger machines get larger models. Smaller machines get smaller ones.
Gives you one API.
Your agents and apps call a single address on your machine.
Routine tasks run on your machine. Jobs that need more power can move to cloud GPUs, with an upfront quote paid in SOL.
02 / Try it
Run a session on this page.
Connect an agent and watch requests run on a 12 GB machine. Then decide what happens to a job that is too big.
This machine · GPU · 12 GB · Coding model · 7B loaded
Request log
localhost:7317No client connected. Connect a coding agent to start the session.
This machine
Cloud spending
0.014 of 0.250 SOL used this month
Cloud access
Auto-approve under
SOL
Monthly cloud limit
SOL
03 / Why run it yourself
Your machine does the routine work.
Hardware you own.
Use hardware you already own instead of paying a provider for every request.
Local when you choose.
Keep selected tasks local, with cloud access disabled if you prefer.
Reach it from anywhere.
Access your home machine from your laptop or phone.
One spending limit.
Set one spending limit for cloud jobs.
04 / Your devices
One machine at home. Every device you own.
Laptop
Using Home desktop · Local
Phone
Using Home desktop · Local
Home desktop
GPU · 12 GB
Only with your approval
Cloud GPUs
Quoted per job · SOL
Your laptop and phone use the models on your home machine.
05 / The API
One endpoint. You decide where it runs.
POST http://localhost:7317/v1/chat
Content-Type: application/json
{
"model": "auto",
"messages": [
{ "role": "user", "content": "Explain this error: undefined is not a function" }
],
"allow_cloud": true,
"max_price_sol": 0.02
}Fits on this machine
{
"ran_on": "local",
"model": "coding-7b",
"tokens_per_s": 43,
"charged_sol": 0
}Too large for this machine, approved for cloud
{
"ran_on": "cloud",
"model": "coding-70b",
"tokens_per_s": 118,
"charged_sol": 0.018
}Works with tools that already speak common chat-API formats.
06 / FAQ
Questions
What does Granio install?
One desktop app. It checks your hardware, downloads models that fit, and runs an API on your machine at localhost:7317.
Which hardware does it support?
Granio checks yours and picks models that fit. Smaller machines get smaller models.
When does a task go to the cloud?
Only when it does not fit your machine and you approve the quote, or when the quote is under your auto-approve limit.
Can I turn the cloud off completely?
Yes. With cloud access off, tasks that do not fit are declined. Tasks set to Local only never leave your machine.
How do I pay for cloud jobs?
Each cloud job shows an upfront quote paid in SOL. Your monthly cloud limit caps what you spend.
Can I use it from my phone?
Yes. Your phone connects to your home machine and uses the models running on it.
How do I get access?
Join the early access list below. We email you when Granio is ready to install.