Trinity product · deliberation companion · pre-launch

sasha

a room that pushes back

Not a chatbot that agrees with you. A room that pushes back: a personal board of AI advisors — Sun Tzu, Marie Curie, Nietzsche, Simone de Beauvoir, Ada Lovelace, and others — built to argue a hard decision out in one of five formats, and end not in a hedge but a verdict: a vote, a majority opinion, a documented dissent, and one next step.

Get early access See it live →

In one line: ChatGPT averages. sasha argues. One synthesized opinion vs. a board that disagrees on purpose, so the strongest case for every side actually gets made.

A Trinity product — part of the venture wing's floor of instruments see the rest of Trinity →

Five formats

№ 1

Supreme Court

a formal opinion

Each advisor states a position for the record; the deliberation closes with a written opinion — majority and dissent both signed.

№ 2

Devil's Advocate

deliberate opposition

One voice is assigned to disagree with the room on purpose, so the strongest counter-case gets made even when the room already agrees.

№ 3

Scientific Method

hypothesis, then evidence

Your question becomes a hypothesis the board tests against evidence and counter-evidence before it will commit to a conclusion.

№ 4

Socratic

a chain of questions

Advisors interrogate the question itself before answering it — likeliest of the five to hand you back a better question than the one you brought.

№ 5

War Council

tactical, contested

Options get laid out and argued hard on trade-offs, the way a room planning a campaign would, until one course of action is chosen.

The roster of advisors reads dynamically, never a hard-coded count. The number was never the point — the argument is.

How it works

One hard question, several ways to hold it.

Bring a real decision, pick a format, and let the board argue it out in that shape — a clean verdict from Supreme Court, a contested set of options from War Council.

The disagreement is the point.

Every deliberation carries a required, equal-time dissent. A unanimous vote isn't a win — it's treated as a bug and re-checked.

A verdict, not a chat.

The output is structured, not a wall of text: per-advisor statements, a vote, a majority opinion, a documented dissent, and one concrete next step — built to be dropped straight into an actual meeting.

Callable from where you already work.

sasha also runs as an MCP server, so the same board can be reached from Claude Code, Cursor, or another agent workflow — not only the web app.

What it won't do

It will never average. Every deliberation carries a required, equal-time dissent — a clean, unanimous vote gets treated as a bug, not a result.
It will never claim to be the historical figures. Every persona is framed as an AI assistant inspired by someone's public philosophy and documented work — never as the person themselves, and never speaking for them.
It will never put a modern fact in a historical mouth. Web research lives in the synthesis, never inside an individual advisor's own statement.
It won't play doctor, lawyer, or therapist. Medical, legal, and financial questions hit a gate before they reach a clinical-sounding answer — a disclaimer alone isn't treated as enough.
It's decision support, not emotional support. No companion framing, no roleplay — the lane is the decision in front of you, not the feelings around it.
It won't engage someone in crisis. The product gates by age, and crisis language is routed to real hotlines before any board convenes.

Live, and not open to everyone yet

sasha is real, working software, reachable today at sasha.stellarii.studio — ahead of the wider Trinity v1.0 rollout. Leave your email and we'll write when broader access opens.

Draft copy — exact pricing and the full rollout sequence publish once locked.