Bring your own AI. We'll run the system around it.
The hosted plan is the Jellyfish Assistant — the same system as the open-source release — run for you, in an encrypted space of your own. You bring the models you already use. You decide which one does what, and what each one may see.
Not a chatbot
You're not buying a chatbot.
A chatbot is one model in a tab. The Jellyfish Assistant is the system around the models: the memory, the tools, the permissions, and the record of what happened. The hosted plan is that system, already set up, on our machines instead of yours. Running it yourself stays free.
What the plan is designed to include
- All first-party software, kept up to date.
- The multi-provider backend. Bring keys for the providers you use, or connect your own Ollama.
- Per-task routing. You choose which model handles which job.
- Data-flow rules. You choose what each model or provider may see — your email for one, your code for another, your memory for none.
- An encrypted space of your own for memory, chats and files, with backups.
- Add-ons, first- and third-party, under the same sandbox and permissions.
- Accounts for your household, with roles, devices and second factors.
- A web app first; mobile apps planned.
Amounts — storage, backups, limits — will be published before anyone is asked to pay.
What it doesn't include
Inference.
You pay your model providers directly. If you send something to a frontier provider, that provider sees what you send; the Jellyfish Assistant's job is to make sure that only happens where you said it may.
What we track
Two things.
How much storage you use, and aggregate resource usage. Not what you store, and not what you say.
What we never do
We will never sell your data.
Not to advertisers, not to anyone.
Later · optional · planned
Private inference, hosted by us.
Private, secure inference for the work you want to keep your own: open-weights models on our machines, with clear usage limits, added to the bring-your-own-key plan if you want it.
News and code can go to the frontier models. Your mail, your calendar and your memory belong on a private channel.
What we can't tell you yet
Some things aren't decided.
Storage amounts, backup frequency, when it opens, and what either plan costs.
Hosting means our machines. Before launch we'll publish exactly what that lets us see and what it doesn't — including what the Jellyfish Assistant's design can't protect against.
Questions
Three answers.
Is the open-source version going away?
No. Self-hosting stays free.
Do I need the hosted plan?
No. It's for people who'd rather not run a server.
Can I use my own Ollama with it?
That's the plan.
Hosted interest
Interested?
No payment and no commitment. This list keeps you informed, and tells us how many people want this.