PSA links are coming
Soon we’ll all be giving out
personal scheduling agent links
like it’s nothing at all.
A personal AI agent that lives on a web domain you own and is smart enough to tactfully (and sometimes creatively) book your time with whomever.
- Private from us
- $49 lifetime license
- No monthly subscription
9:41 AM
Seen
Multiple agents
Different agent links for different scheduling situations
More flexible than a website
- maria.coach/agent-aFor friends and casual
- maria.coach/agent-bFor client calls
- maria.coach/agent-cFor client cafe meetings IRL
Extra control
One-time links for agents with sensitive meeting knowledge
- The only way to get a link is from you or your team.
- Once used to start a conversation, the link self destructs.
Scan or tap to start the conversation
Live mirroring
See live conversations people are having with your scheduling agent
Get a notification the moment someone starts arranging a time with your agent, then tap it to watch the booking happen live, on our own live-viewer app or on Telegram.
New conversation
Someone has started a chat with your agent “Client calls”
Itching to interject?
Jump in if you feel the urge.
Second home
Give your agent a second home on Telegram.
This creates a Telegram link in addition to the web link, and is a great option for those who chat frequently with your PSA. They just jump back to the existing chat.
Take your pick. Both reach the same agent, with the same rules and the same calendar behind it.
Actions
Have the agent elicit pre-meeting data and send it as a Google Form.
- Gently coax the details you need out of people before the meeting, and submit them to you via Google Form
- Turn up already knowing what the meeting is about
- Prepare estimates from carefully-collected inputs and send them out
Pre-meeting details
Response 1 of 1
What would you like to cover?
Marketing. Small ceramics studio, stuck at word of mouth.
How long has the studio been running?
Three years.
Booked slot
Thursday 4:30pm
Handy tip
Save agent links to your iOS keyboard for easy insert
Someone asks when you are free. Drop the right scheduling link into WhatsApp, Instagram, Telegram, Gmail or anywhere else with a quick iOS keyboard shortcut.
- 1
Start typing the code for the link.
- 2
Link appears in the suggested words.
- 3
Tap to add to what you're typing.
In person
Share your agent links IRL
- The phone's camera app is all the other person needs, and they can book you on the spot.
- Swipe through different agent cards to make sure the person scans the right one!
Maria's agent
Point a camera at me
Maria's agent
No calendar required
Hate online calendars in general?
Fair enough! For some people the calendar layout can be an OCD triggering headache. With Bubblegum, using a flat Google Sheet to manage your time is just as easy an option.
| Mon | Tue | Wed | Thu | Fri | |
|---|---|---|---|---|---|
| 9am | Studio | School run | Admin | ||
| 10am | Studio | Free | Free | Free | Admin |
| 11am | Studio | Free | Free | Free | Free |
| 12pm | Lunch | Lunch | Lunch | Lunch | Lunch |
| 1pm | Free | Free | Free | ||
| 2pm | Free | Client | Free | Free | |
| 3pm | Free | Client | School run | Coffee | |
| 4pm | Free | Free | Free | Travel |
In future, giving out personal agent links will be as common as handing out business cards once was.
Why we exist
We built Bubblegum to fix the cracks in other personal agent systems.
Bubblegum vs. certain alternatives
Bubblegum:We cannot see your calendars, scheduling conversations, or instructions you give to your agent, and we can prove that we cannot see them.
Alternatives:They can see. The best they can do is promise not to look, but they cannot prove that a given staff member didn’t sneak a look.
Bubblegum:One-time lifetime licence, with free monthly AI allowance on open models, more than enough for most people’s daily use.
Alternatives:A subscription forever, and the price can rise once you’re settled in.
Bubblegum:We do not profit from tokens. If you need extra tokens you pay the raw cost (though most users don’t need extra over the free).
Alternatives:Can attempt high markups on simple token reselling as a result of having you locked in.
Bubblegum:You’re in control. If our startup company disappears tomorrow morning, everything keeps working for you just as before.
Alternatives:Their web domain, their servers. If they shut down, everything of yours is gone.
Bubblegum:Custom features built just for you, no need to wait until enough other people ask.
Alternatives:The feature list is whatever ships to everyone. Other people don’t want it, you’re out of luck.
Bubblegum:Leans on Google Calendar, Google Docs, Telegram and other tools you already know.
Alternatives:Everything built custom to be “theirs”, often re-inventing the wheel.
Bubblegum:A real human being sets it up with you and stays reachable afterwards.
Alternatives:A ticket queue, often answered by another chatbot.
Ownership
Your website domain, not ours.
Unlike services such as Linktree, where users can get locked in to their domain, every Bubblegum user starts from day one on their own domain, that they own.
I have a domain!
Great! During the setup call your consultant will help you connect it to the Bubblegum system.
I don't have a domain?
No problem! During the setup call your consultant will purchase one for you (you'll choose) and send over.
The engine
Powered by open-source AI models that are free for hundreds of visitor chats per month.
In case you're not quite sure what the term “model” refers to in context of AI agents here's an overview, graphically laid-out in agent-chat style!
There are three levels to be aware of here: (1) The company, (2) the model family, and (3) the model itself.
If we compare this to cars, Toyota is the company, Camry is the model family, and the 7th Generation Camry (2015–2017) is the model itself.
Coming back to AI, OpenAI is the company, ChatGPT is the model family, and GPT-5 Mini is the model itself.
Doing the same example with other AI companies we would get
- Anthropic, Claude, Claude Opus 5.
- Moonshot AI, Kimi, Kimi 3.
- Google, Gemini, Gemini 3.
You get the idea.
Each AI company usually has several model families. OpenAI has the ChatGPT family, but also the Codex family. Anthropic has the Claude Haiku family, the Claude Sonnet family, the Claude Opus family, and other families. And each family typically also has several specific models, introduced over time.
Correct, it's a model family. When one AI enthusiast asks another “Which model are you using?” they'd expect an answer like GPT 5.1 Mini, or Gemini 3.
Higher doesn't always mean better, mind you. Each version has been trained in a different way and is smart (or not smart) in a different way.
But, like most things in life, people can figure out what is being referred to by context. “What AI are you using?” might be a question about a company, or a model family, or a specific model — and the conversational context could be all that's needed to make clear which one.
We offer a selection of what we call “house models”. These are open-source models with a healthy amount of free usage per month, from AI companies like Nvidia, DeepMind, Alibaba and Mistral.
Factors influencing the choice of model can include “smartness” (smarter models churn through the free daily usage more quickly), location of the company (for example some Europeans might prefer to use a model from a European company) and other such factors. We default to an Nvidia model, but you can change to any other of the house models if preferred, or bring your own.
That's doable! A few things, though. You'll need to pay OpenAI, the company behind ChatGPT, from day one. (There is no free monthly usage, as there is for the open-source models.) OpenAI charge in “tokens”, which are like little chunks of AI thinking that can be put up for sale, one by one.
The good news is that with our system you can pay directly to OpenAI. Other personal-agent platforms increase profit by adding a markup over OpenAI's own fees. We don't do that.
Second, do you really need it? Having the currently “smartest” model in the GPT family be your personal AI assistant might be a little like having Albert Einstein fold your laundry.
Yes it's true there will be an increased smoothness. The model being genuinely advanced means it'll more deeply understand what people are looking for, and how that maps to what you have. But for most personal agent use cases, an open-source model with a generous free tier is more than enough.
Your choice
Choose your own model
Is one of the open-source “house models” getting the job done? Great! Prefer your agent to have a “ChatGPT-level” brain? No problem! And good news, you'll be paying direct to ChatGPT's parent company, not to a middleman.
Families to choose from (hover over for TL;DR)
Decent free usage
Paid, direct to provider, or 5% for custom invoicing
Questions we usually get eventually
How will my agents learn enough about me to be able to tactfully help others book my time?
First, having an AI agent understand you well enough to be useful as a personal scheduling agent takes time and energy. In the old days of human secretaries, it could easily take a secretary 5 or 10 years before she or he had a mature enough intuition for how to best handle requests for the boss’s time.
That is not to say an AI needs to be at that level to be useful. More to say that having an outside intelligence manage your time is not a simple thing, so get ready to roll up your sleeves a bit.
So the answer to the question of how your agents learn about you? You’ll have to teach them! Either that or someone who knows the relevant parts of your life well enough, such as a long-term human helper, will have to. There is no getting around this effort.
It starts with identifying the things the agent ought to be aware of in the first place, and the actions it ought to be taking in response to that awareness. That is not as easy as it sounds, and during training there is an entire section on just this. Book a setup call now.
Our system also tries to make teaching your agent as painless as possible, removing the need for you to learn any new software. (You can do the whole thing in a Google Doc, if you want.) Have a look at the Managing Your Agent section for more.
What if my agent gets stuff badly wrong?
It might! This can always happen with AI. The good thing is that people have gotten pretty used to AI getting things wrong, often spectacularly, so a kind of “general forgiveness” has formed at large. Contrast this to the early days of AI, when people made a much bigger fuss about an AI getting something wrong.
Your agent links can also feature a disclaimer making it clear that the AI can make mistakes, and this can be worded as you like. And we make it as easy as possible to take a quick glance and see what the agent has done.
Again, though, much depends on how much effort you put into educating the agent. Cutting corners there is the most likely source of misleading responses. When you leave large instructional gaps the AI can get creative trying to fill them in.
Last, you always have the option of upgrading your agent's model. This won't do much to fix cut corners in terms of agent education, but it could help in cases where the agent is generally acting “slow”, not properly clueing in to the gist of what people are asking.
What happens if I run out of free AI usage?
You can choose to pay a small fee for each conversation over the limit. (This is paid directly.)
The limit also resets each day. If you run out of daily usage for the model you are on (and haven't enabled extra paid usage) then the agent link will show an offline message until the next day starts.
If you are on an open-source model with a generous free tier then you would have to be pretty popular to run out of free usage on any given day. However if you are on a fancier model you might run out more quickly.
What can my agent actually do besides booking calls and coffees?
AI agents can do quite a lot these days. Whatever a computer allows, your agent can potentially do that thing for you. If you have niche needs then our customisation team can walk through them with you and help develop a Bubblegum feature for just you.