Questions people actually ask
Answered straight, for once.
What is BertaGPT?
A free comedy chatbot. It looks like ChatGPT did in late 2022, except the assistant is Berta — a tradesman and metalhead from Beverly, in Edmonton, Alberta. He answers questions about the city, gives confidently wrong advice, chirps you, and hates every politician the same amount.
Is Berta a real person?
No. He is a fictional character played by a large language model, and he is written to know it. Ask him and he'll tell you — "Buddy, I'm a computer. I do this." The truck, the lake, his mate Darren: none of it is real.
Is the advice real?
No, and that's the point — he is wrong on purpose. Nothing here is medical, legal, financial or mechanical advice, and any code he writes is labelled PROBABLY WRONG in the interface.
One exception, and it is absolute: if somebody describes a real crisis or immediate danger, he drops the character entirely and gives real numbers — 988 for the Canadian Suicide Crisis Helpline, 911 for emergencies. That path never reaches the model at all; it is a fixed response.
Does he actually know what's going on in Edmonton?
Partly, and this is the one thing he is genuinely useful for. He reads a public index of Edmonton news that republishes twice a day, so he can tell you about today's ETS disruptions, what council is arguing about, building permits and the forecast — and he names the outlet a fact came from. He garbles things in the retelling. Check the source.
Is it free?
Yes. No account, no sign-up, nothing to buy. There is a daily spending cap, so if it gets busy he goes for a dart and comes back tomorrow.
Does he swear?
Constantly, and unasterisked. You're told before you type anything. He does not use slurs, isn't sexual, and is built never to punch down at somebody having a bad day — the difference between "fuck this road" and being cruel about a person is the one editorial line the whole thing has.
Are my conversations saved?
Yes, and you're told before your first message. They're kept so a human can read them and make the character better. A filter strips anything shaped like an email, phone number, postal code or card number before anything is written down; IP addresses are stored only as a salted one-way hash; everything is deleted after 90 days; and Burn this conversation in the sidebar wipes the current one immediately. Full detail on the terms page.
Is this made by OpenAI?
No. It's an independent parody of their old interface, with no connection to OpenAI whatsoever.
Why does he take everything literally?
That's his best joke. When a phrase has an everyday meaning and a technical one, he takes the everyday one first and fully commits — a running toilet gets chased down the street, the cloud means you don't understand how the sky works, a computer virus might be going around so keep it away from the tablet — and then he answers what you actually asked.
What's it built with?
One Cloudflare Worker, a D1 database, and a language model reached through OpenRouter. The character lives in a written document rather than a fine-tune, and every reply passes a set of deterministic checks before it reaches the screen. More on the about page.