August 22, 2026 · 6 min
The private ChatGPT alternative you actually own
People go looking for a private ChatGPT for a few different reasons, and they don’t all want the same thing. Some want their chats out of a company’s account. Some want an assistant that remembers them without that memory sitting on someone else’s server. Some just want to stop paying per seat for something they use twice a week. This is a straight account of where meraGPT fits those, and where it doesn’t.
What “private” can and can’t mean here
Be honest about the wire first. When your agent answers, the prompt goes to whatever LLM your API key points at. If that key is an OpenAI key, the text reaches OpenAI. No product that calls a hosted model can tell you otherwise, and the ones that imply they can are selling you something.
What changes is everything around that call. Your conversation history and your agent’s memory live on your own box, in a file only your box can read. They are not in a ChatGPT account tied to your email. The big API providers also don’t train on API traffic the way a consumer chat app trains on chat logs. So you get the privacy people actually mean when they say it: my history is mine, my data isn’t the product, nobody is building a profile out of my Tuesday afternoon questions.
If you need the model itself to run with no network at all, that’s a fully local setup, and it’s a real thing you can want. meraGPT is a private box around a hosted model, which is a different point on the same line and worth being clear about.
What you give up coming from ChatGPT
I’d rather you know this before you sign up than after.
- You bring a key. There’s a one-time step where you paste an API key from a provider. It takes a couple of minutes and then you forget about it, but it’s a step ChatGPT doesn’t ask of you.
- The default model is a cheap one. We start you on Gemini Flash Lite because it’s fast and costs almost nothing. If you want GPT-5-class answers, point your key at that model and pay that provider’s rate. You choose per message, which is the upside. It does mean the out-of-the-box model isn’t the smartest one on the market.
- It’s an agent you reach from a web chat, or Telegram, or Slack. There’s no iOS app with voice mode and a hundred million users’ worth of polish behind it.
What you get that ChatGPT doesn’t give you
- Memory you own. The agent keeps a private memory of who you are and what you’re working on, and it sits on your box. You can read it, edit it, and take it with you.
- It does things on a schedule. A cron job can have it read your calendar at 7am and message you the day, or summarise your inbox before you open it. A chat window waits for you; this doesn’t have to.
- It lives where you already are. Wire it into Telegram or Slack and you talk to it from an app that’s already open on your phone, no separate tab to remember.
- Costs track usage. You pay the box fee plus what your model calls actually cost. A light month is a cheap month. No per-seat plan sized for someone who uses it all day.
Who this is for
It fits if you want an assistant that’s yours, that remembers you, and that you can reach from the tools you already use, and you’re fine bringing a key to get there. It fits if the privacy you care about is ownership of your history rather than an air-gapped model.
It’s the wrong tool if you need a frontier model in a finished consumer app and don’t care where the data lives, or if your requirement is a model that runs with the network unplugged. Both are reasonable things to need. This isn’t them.
Trying it
It’s $9.99 a month for the box, your own key for the model calls, and about three minutes from sign-up to an agent you can talk to. If that’s the shape you were looking for, the quickstart walks the whole way through, and the piece on what “private by default” means goes deeper on the ownership side.