| Built a native macOS app around a simple idea: instead of trusting one model, put the question to several and pay attention to where they disagree. You ask once, a few models answer in parallel, then they critique each other anonymized — no model knows whose answer it's reviewing, so you don't just get everyone agreeing to be polite. The app then surfaces the real fault lines and writes a synthesis. The disagreement is the interesting part — that's the whole premise. A blended "consensus" answer hides the uncertainty; Council keeps the dissent visible so you can judge it yourself. Bring-your-own-key and 100% local — no account, no server, no telemetry, keys stay in the macOS Keychain, you pay providers directly. Free and open source (MIT). Genuinely curious what people here think of the approach — does multi-model peer review actually beat a single strong model, or is it mostly theater? [link] [comments] |