Get a room full of expert advisors for any big decision — before you commit.
Several leading AI advisors each examine the same thing — your plan, your decision, your draft, your design — then debate it out loud and converge on one clear recommendation. Think of it as the smartest meeting you’ll ever call: everyone did the reading, no one’s afraid to disagree, and you get a clean summary at the end. It works for software, but also for product, research, legal, business, and writing decisions.
Founders weighing a big bet. Writers who want a hard read before they hit send. Product managers stress-testing a launch plan. Researchers checking a method. People comparing two real-life options. And yes — it goes deep for engineers too, citing exact path:line evidence in your codebase. If it’s a decision worth getting right, it’s worth bringing to the board.
path:line evidence.)Here’s a real board, debating a real decision — “Should I go full-time on my side project?”
Verdict: Proceed with care — unanimous, high confidence
Workable, but address the flagged concerns before you go ahead.
The blocker the board converged on: “Income case never closes: $12k MRR ≠ replacing a $165k salary.” All three advisors agreed the financial model was the gating item — once health insurance, self-employment tax, and lost benefits are stripped out, the target income is closer to half of current total comp. They didn’t say “don’t do it.” They said here’s exactly what has to be true first, and handed back the next steps to get there.
Read the full handoff → — a self-contained page that opens offline: the verdict, the round-by-round debate, the blockers they agreed on, the dissent they preserved, and the concrete next actions.
The default board uses frontier models from Anthropic, OpenAI, Google, and xAI — Claude, Codex, Gemini, and Grok — each in a different seat. Provider-maintained selectors resolve the strongest current model at run time. Two rounds of back-and-forth are the default: enough to genuinely stress-test your thinking without dragging on.
Independent first, together second. Each advisor reviews your material on its own, with no idea what the others said — so they catch each other’s blind spots instead of nodding along.
You approve what leaves your computer. You see exactly what would be sent to each provider and approve it before anything leaves your machine — redact what’s sensitive, or run a fully local board where nothing is sent at all.
It looks before it acts. The board only reads, and never changes your work unless you explicitly ask. No surprises, no stored secrets.
A preflight check confirms at least two advisors are healthy before any run begins — a board needs a quorum.
A run hands you a single, plain-English handoff you can read like a memo, plus the full record behind it:
Curious what a finished run looks like first? Open the sample handoff →
Under the warm summary is a workflow built for technical rigor — and a machine-readable verdict you can wire into a pipeline.
Point the board at a real codebase with --repo and advisors cite exact path:line evidence, working from a captured snapshot rather than guesswork. Ready-made lens sets for software, product, research, legal, business, and writing live in lens-presets.md. For software and technical work the seats take distinct lenses:
opus, latest-Opus alias) — architecture, systems, and adversarial design review.pro, highest-reasoning Pro alias) — product, operations, rollout, latency, evaluation, and user-workflow risk.grok-4.5) — contrarian synthesis, hidden assumptions, alternatives, and decision-changing evidence.Every seat still reviews the whole brief; the lens reduces blind spots, it doesn’t narrow responsibility.
Two rounds is the default. Set rounds to auto to stop early on convergence or add a round when real disagreement remains.
Every run also emits verdict.json — a machine-readable call (ship | caution | block) you can gate CI on or reshape into a PR comment, Slack message, or TL;DR. scripts/board_verdict.py --gate turns the board’s call into a CI exit code; scripts/format_output.py reshapes it for wherever your team reads it.
What “verified” means here: each cited line carries a resolution check that confirms the quote exists in the captured material — it does not prove the inference drawn from it is correct. The board sharpens your judgment; it doesn’t replace it.
A run saves:
verdict.json — the machine-readable verdict described above;Prefer to see the board on code? The payments idempotency review walks a real architecture decision end to end, with path:line evidence and a unanimous block verdict. Browse the full run in examples/payments-idempotency-review/.