AI that never leaves your desk.
Ask it as much as you like, about your own files, on your own computer. No subscription, no message limits, nothing uploaded.
14 days to try it, then from $8 a month.
Three steps, then it is yours
- 1
Install it
One click, then sign in. Nothing to configure and nothing to set up.
- 2
Pick one it says will run well
It has already checked your computer. The download is about the size of a film.
- 3
Start typing
That is the whole setup. Turn your internet off afterwards if you like.

For individual users
Everything, as often as you like
Ask it anything, all day
No meter and no message cap. The subscription is flat, because the computing happens on your machine rather than on servers someone has to pay for.
Nothing leaves the machine
Not the question, not the answer, not the document you opened. Your account exists to handle the subscription and nothing else — it never sees what you type.
It reads your own files
Point it at a folder of notes or contracts and ask. It tells you which file each answer came from.
Nothing to set up
One installer, then a sign-in. No command line, no Python, no API keys to paste. It works out your hardware for you.
For small practices
When the answer to “where did the data go?” has to be nowhere
A solicitor, an accountant, a clinician or a therapist often cannot send client material to a third-party service at all — whatever it costs, and whatever the provider promises.
The files you cannot upload
Client records, case notes, patient files. Point it at the folder and ask about what is inside — the documents never leave the machine in your office.
Nothing to put in a data agreement
There is no processor to name and no transfer to disclose, because nothing is transmitted. The question your compliance policy asks does not arise.
The models are yours to keep
Specialist models bought from the store are a one-off purchase, not a rental. They keep working on the machines you installed them on.
Works on a machine with no internet
Once a model is installed it never needs the network again. That suits an isolated workstation, and it is the simplest possible answer to where the data went.
For developers
Opening soonSell the models you already make
If you fine-tune models or build adapters, the store will be somewhere to sell them — to people whose machines have already been checked against what you built.
Keep 75% of what you sell
Higher than the app stores take, and affordable here because we host catalog files rather than GPUs. The infrastructure cost of your work is close to nothing.
Charge for work you give away today
A fine-tune uploaded to a model hub earns nothing however good it is. There is no checkout on those sites, and building one yourself means payments, licensing and piracy.
We handle payment and licensing
You upload the file and set a price. Purchases, licence keys and updates are ours to run — and licences verify on the buyer's machine without phoning home.
Only shown to people it will run on
Every listing is checked against the buyer's actual hardware first. Nobody pays for a model their computer cannot load, which is the usual source of refunds and one-star reviews.
The creator side is not open yet. This is what it will be when it is, and the terms above are the ones we intend to launch with.
Honestly, against the cloud
It wins some of these and loses others. Better you know which before downloading five gigabytes.
| Feature | Cloud AI | Bobbin |
|---|---|---|
| How much you can ask | Capped, then throttled | As much as you like |
| Where your files go | Uploaded to a server | Stay on your computer |
| Works offline | No | Yes |
| Where the computing happens | Their servers | Your computer |
| Whether it will run on your PC | Not a concern — it is their hardware | Checked and told plainly, before you download |
| Answer quality | Best available | Good, not the best |
| Speed | Fast, always | Depends on your PC |
The last two rows are the trade. Models that fit on one computer are smaller than the ones in a data centre. For drafting, summarising and questions about your own documents the gap is small; for hard reasoning it is real.