Stop asking your AI to act like an expert. Give it judgment from one.
Connect a Decision Model from a professional whose experience you trust. Your agent reads it before it answers you, in the tool you already work in.
- Claude app
- ChatGPT
- Cursor
- Claude Code
- Gemini CLI
- and more
What it looks like when it is working
Nothing changes about how you work. You ask your own AI, in your own tool; it consults the model first and answers with that judgment already in hand.
Example inside your AI tool
We're a 20-person Series A. Our marketing site looks dated next to our two biggest competitors and the board keeps bringing it up. Do we redesign the brand or the site first?
Consulting Ines’s Decision Model
- Recommendation
- The site, and only the site. Keep the name, the mark and the wordmark exactly where they are. New typography, new photography, and a homepage headline that says what you do and for whom.
- Why
- “Dated” is almost never an identity problem. What a board is reacting to is the typography and the fact that your headline does not separate you from the two companies beside it — both of which live in the pages, not in the mark. Identity is the output of positioning: change it before the positioning is settled and you will change it again, having spent the one design budget you had on the wrong problem.
- What would change this
- If the company has actually moved — new segment, new buyer, a different room — then the mark is fighting the new positioning and the brand goes first, site second. That is a change in audience, not a change in taste, and it is the only version of this question where the logo is the answer.
- Relevant precedent
- Seed-stage B2B startup, 11 people, 40 paying customers, no clear positioning yet. The founder wanted a full rebrand — new name, new logo, new site — before the next raise. → No rebrand. Fix the homepage headline and the product's empty states; leave the name and mark alone.
- Outside this model
- Nothing here about what a redesign should cost, who should build it, or how to run the agency relationship. This model covers what to change and what to leave alone.
Ines’s own answer, on a question they chose. Your AI writes the wording for your situation; the judgment is theirs.
Ines is a reference model — a composite avva wrote, not a real person, so the answer above is ours to publish. A real expert’s page carries their own.
See the technical outputWhat the connector actually returns, for anyone wiring this up
Each tool call returns a verification packet: the published Decision Model plus a procedure, never a finished answer. Your own model does the reasoning.
- Their decisions
- The calls they actually made, with what each one turned on. At full access, ordered against your situation.
- Forces and their order
- The pressures in play and which wins when two collide — read off the decisions, never ranked by hand.
- Stop rules and switches
- The lines they will not cross, checked first, and the conditions under which their usual call flips.
- An output format
- Including a line for what the model was silent on, so silence is never returned as approval.
Five tools: review before finalizing, red_flag_check as a cheap gate, expert_questions before an approach is locked, compare_options for A/B, about_expert for scope.
Inside a published model
Your AI can argue either side of a trade-off equally well, so which side you get depends on how you asked. A Decision Model settles that in advance, the way one person settled it: not a questionnaire and not a chatbot wearing their name, but a small map fitted from decisions they really made — the decisions themselves, what each one turned on, and which pressure wins when two collide. Nobody fills in a ranking; it is read off the decisions.
- vs. a persona
- “Act like a senior architect” produces a costume: the average of the internet, in a voice. This is one named person’s real calls, with what each turned on and what they refuse.
- vs. a knowledge base
- A document says what somebody knows. A model says how they decide, and your AI applies it to a situation the expert never saw.
- vs. hiring them
- Nobody is notified and nobody is on call. It is a published record your AI reads at the moment you are deciding, on the calls too small to book someone for.
Connecting takes one line
If you can paste a link into your AI tool’s settings, you can do this.
Choose a Decision Model
Open a professional’s page and read what they weigh, where they stop, and some of the decisions behind it — before you connect anything.
Try it free
Copy the connector from their page into your AI’s settings. No account and nothing to install: it returns a shorter slice and 40 consults a day, per model, which is enough to see how they reason and whether it fits your problem.
Keep it for ongoing use
Sign in for the complete published model, then paste their standing note into your project instructions — and your AI consults them on its own from then on.
AI apps
Paste a link in settings.
- Claude app
- ChatGPT
Coding agents
One line in the project config, or one command.
- Cursor
- Antigravity
- Claude Code
- Gemini CLI
Developer integration
Wire a model into something you are building.
- Agents SDK
The exact line is on each model’s page: it carries that model’s own address, in the shape your tool wants, with a copy button and a message to send afterwards so you can check it worked.
Whose AI answers
Yours does. avva runs no AI.
The expert builds their model with the assistant they already trust. You apply it with the AI you already pay for. Nothing of ours sits in the middle, so nobody swaps in a cheaper model behind your back and there is no second subscription — and the quality of the reasoning improves when your own model does.
- Their AIreads the expert’s own work history and drafts the model. They approve every decision in it.
- avvastores and serves the finished Decision Model. It writes no answers and runs no model of its own.
- Your AIreads it before it answers you, and names where the expert had nothing to say.
What it does not claim
- No leaked Slack, no client decks. avva never receives the expert’s source material. What publishes is their calls, the forces behind them and their rules — client names, employer names and source quotes are stripped before storage.
- No expert hiding in a server. Nobody is notified, nobody is on call, and nothing you ask reaches a human.
- No accuracy claim. Coverage labels — Light, Solid, Broad — describe how much approved material a model rests on. avva publishes no score predicting how often anybody is right.
- No silent guessing. A model is one person’s slice of a field, not its boundary. Every answer names what it had no position on, and where it is silent your AI says the judgment is its own.
Where the order inside a model comes fromFor anyone who wants to know that nobody types the numbers in
One decision reveals several contests
Each case records which pressures settled it and which lost. That split alone is a set of head-to-head contests. One decision gives four here, and nobody ranked anything.
Pressures get a resolution order
The order comes from the method tournament statistics uses to rank players from their match results, run on those contests. It answers the one question that changes what an AI does: when two of this person’s pressures collide, which wins.
Which case counts as nearest
A case built on pressures you do not have is about something else, so it is penalized in full. Extra pressures on your side cost little. The shared ones still hold.
The rest runs on hard minimums: every published rule carries the count of approved decisions behind it, a hard stop needs at least two, and one decision becomes at most one case. What crosses which line is in the Privacy Center.
Free during beta
The trial needs no account and no card. The reasoning runs on the AI subscription you already pay for; avva adds no per-call charge.
Publishing rather than connecting? What it takes, and what stays yours. Buying for a team? A private shelf.