What avva is, and how it works

What is avva?

A place where a professional turns how they decide into a Decision Model, and where other people connect that model to their own AI, so the AI checks it before it answers in that person’s field.

What is a Decision Model?

A professional’s past decisions, kept as cases: the situation, what they decided, and why. Around the cases, the forces they weigh and the order those resolve in, where they stop, and what they ask before deciding. It stands in for the “You are a senior role who…” prompt a reader would otherwise write for their AI. It never answers as the person.

Does avva run an AI?

No. Two AIs are involved and neither is ours. The expert’s own assistant reads their history and drafts the model. The reader’s own AI reads the model and does the reasoning. avva stores the finished model and serves it.

What does MCP mean?

Model Context Protocol, the open standard AI apps use to reach tools outside themselves. Every model on avva is an MCP server at its own address. Connecting means pasting that address into your AI app; the app does the rest.

Why not just write a good prompt?

A prompt written from a job title gives your AI the textbook of a field, which it already has. A model gives it the decisions one person made, the reason behind each, and where that person departs from the usual answer. That is the part a stranger cannot write down for you.

What is avva not?

Not a chatbot that imitates the person, and not a consultation: nobody is notified, nobody is on call, and nothing you ask reaches a human. Not a knowledge base or a dump of someone’s documents. Not an accuracy guarantee, and not a replacement for a human’s accountability.

What do I need?

An AI app that supports MCP: Claude app, ChatGPT, Cursor, Antigravity and Claude Code, or any other client that can add an MCP server. No account to try a model.

What happens when my AI consults a model?

Your AI sends the situation it is working on. avva answers with the relevant part of the model, with instructions for applying it. Your AI reads it and answers you in its own words. The expert is not involved and is not told.

What does a trial give me, and what does a key add?

Without an account, a model answers with a smaller slice of itself, up to 40 requests a day, per model. Sign in and take a key, and your AI reads the complete published model, up to 1,000 requests a day, per model. A key stops after 90 days; take a new one on the model’s page.

How much of my AI's context does a consult use?

A full answer from a model with 60 cases is about 20,000 tokens; at the 200-case maximum it is about 57,000. Without an account the slice is smaller. Your AI decides when to consult; the note you add tells it to skip trivial changes.

Is the answer advice?

No. Your AI’s answer is your AI’s. A model is not advice from its expert and creates no professional relationship with them; the Terms say the same.

What does avva learn about what I ask?

What your AI sends is used to answer that request and is not kept. avva counts requests, so the daily allowance can be measured and a model’s owner can see whether anyone used it; the counts hold neither your question nor the answer. The Privacy Center has the detail.

What is a demo model?

A composite written by avva as a worked example, marked Demo wherever it appears. Its expert is invented. It answers in full with or without a key, so you can see what a connected answer looks like before a real expert in that field publishes.

Can I get the model as a file?

No. Your AI reads it through the connection, one request at a time, and a key is for that reading. The complete model is what a key opens; it is not something to take away.

Browse Decision Models

What does it take?

The AI you already use, with avva Studio added to it. Your AI looks through the decisions you have already made, shows you every candidate in full for your approval, drafts the model from the ones you approve, and reads the draft back to you before anything publishes. You need an account only at the publish step.

What if my AI cannot see my work?

Attach the memos, briefs and reviews it should read. You approve the list of records the same way whatever the source.

What does avva receive, and what does it keep?

Your sources never reach avva. The decisions you approve pass through Studio and are kept only as the cases of your model: as a draft for up to 7 days, then as the published model until you delete it. Source quotes and private examples used only to check the model are removed. The Privacy Center walks through every step with what is held at each.

Can I edit a published model?

Not on the site. Corrections go to your AI before you publish, while it reads the draft back to you; a change of mind later is rebuilt from the file your AI saved on your side, published as an update to the same address. The one thing the dashboard changes in a model is which cases are shown first.

Who can see it?

Your choice, changeable at any time: in the marketplace, by link only, private to you, or on a workspace for the colleagues you name. A new model is private until you choose otherwise. Opening a private or workspace model to everyone asks you to confirm privacy again.

Can I take it down?

Unpublish or delete it from your dashboard. The page and the address stop answering; the address stays reserved to you and is never given to someone else. What a reader’s AI already read stays with them.

Can someone copy my model?

Anyone whose AI has access reads the published model in full, and an AI that read a case keeps it. What you publish is what you approved for that audience; private and workspace modes keep it to the people you named.

What does it cost, and will I earn from it?

Free during beta, with no card. Building the model runs on the AI subscription you already pay for. Paid access is not open yet; when it opens, you decide whether your model is ever paid.

Build your Decision Model

What is a workspace?

A private shelf of Decision Models for one company. The person who creates it is its owner and keeps the member list; each member takes their own key, and the key stops working on their next request once they are removed. To anyone not on the list, the shelf does not exist.

Who owns a model on the shelf?

The colleague who published it. It stays on the shelf and keeps answering when they move teams or leave; they can take it down or move it to private on their own. There is no ownership transfer yet.

Can a workspace model go public?

Only once the workspace owner allows it, per model, on the workspace page. Until then a move to the marketplace or to a public link is refused, and the owner is emailed whenever a model leaves the shelf.

Who is behind this?

avva is a free beta run by its founder, Kirill Dobrinov, with the n1.team studio. There is no company behind it yet. Nobody can sign a contract or a data-processing agreement on avva’s behalf until a legal entity exists; when one does, the Terms will name it. If your procurement needs one before then, write to support@avva.chat and we will say where that stands.

Where does my data live?

On Netlify, in the United States, which hosts the site, its functions, storage and database. Sign-in and the visit counter are Google’s; email delivery is Mailtrap’s, in the European Union. Nobody else receives it, and avva sells nothing to anyone. The Privacy Center names each and what to ask of us.

How do I reach a person?

hello@avva.chat for anything, support@avva.chat for an account, a key or a model. A person reads both.

Is any of it open?

The plugins that connect an AI to avva are public, at github.com/kdobrinov/avva_advisor. The site and the server are not.