Ask your AI agent what a real expert would do.
Professionals turn their past decisions into a model your AI checks before it answers. You see what they did, and why.
- Claude app
- ChatGPT
- Cursor
- Claude Code
- and more
Why not just write a prompt?
You’ve probably tried one of these:
- Custom instructions that start “Act as a senior…”
- A paragraph of context, pasted into every new chat.
- One more correction, when it confidently suggests something you would never do.
A prompt can only carry a decision you have already made. For anything new, you’re writing down a guess, and your AI follows it to the letter.
A Decision Model is a record of what one professional actually decided, case by case, with their reasons. The cases show how they weigh a trade-off and where they stop, and which way they went when two good goals collided. Your AI applies that to situations the expert never saw. When the expert adds decisions, your AI reads the new version the next time it asks.
See it answer a question
You ask in your own AI app. It checks the expert’s model first, then answers.
How avva works
- An expert builds a model
Their own AI drafts it from decisions they have already made. They approve every decision in it before anything is sent.
- You add it to your AI, once
Add one link to Claude, ChatGPT, or Cursor, and a short note that tells your AI when to check it.
- Your AI checks it first
Before it answers, it reads the expert’s model. It tells you where they would push back, and which of their past decisions it’s leaning on. If they never faced anything like it, it says so.
When you would reach for an expert
When a mistake would cost real money or weeks of work, and you need the reasons as well as the answer.
Before you commit to an approach
Nobody senior is around to check it. Your AI flags what the expert would not sign off on, and why.
When the room disagrees
Two good options, each with a real argument behind it. You see how the expert’s own cases went when those two goals pulled against each other.
When you are new to the problem
You don’t know what to ask yet. Start with the questions the expert settles first, before anyone has an opinion.
Ask your senior people without booking them
The same few people get pulled into every hard decision. Put their models in a private workspace, and your team’s AI checks them before anyone books a meeting.
- Only invited colleagues can open it.
- Never in the marketplace or in search: to anyone outside, it looks as if it does not exist.
- Remove someone and their access stops on their next request.
In the marketplace
One professional per model, with the decisions they made and their reasons. Composites avva wrote are labeled Demo.
Be the expert their AI asks
You answer the same questions over and over, one person at a time. Publish your model once, and it becomes a link you hand to clients. Their AI checks how you decide before it advises them, and the ones who need more than the model book you through your page.
Soon Earn from it when paid access opens. You decide whether your model is ever paid.
What this is not
It isn’t a chatbot of the expert
Nobody is cloned or impersonated. Your AI reads a record the expert published.
The expert isn’t on call
They aren’t notified, and nothing you ask reaches a person.
We don’t run an AI
When your AI asks, we send the expert’s model and short instructions for using it.
Coverage isn’t accuracy
Coverage labels describe breadth: how much approved material a model rests on. We publish no score that predicts how often someone is right.
You can’t download it
Your AI reads it one request at a time, and we count the requests.
Everything in a model was approved by its expert, with client and employer names already taken out. Details in the Privacy Center.