One question. A roundtable of AI minds. You decide.
A native macOS app that puts the same question to several LLMs, lets them
critique each other blind, and shows you where they agree — and where they don't.
Asking one model a hard question gives you one model's blind spots. Council convenes a panel instead. The same prompt goes to several advisors at once — Claude, GPT, Gemini and more — each answers independently, and then the interesting part begins.
- Ask. Pose a question to your council (three advisors; each seat can be any of twelve backends).
- Parallel answers. Every advisor responds at once, streaming live in its own panel.
- Blind peer review. Each advisor critiques the others' answers without knowing who wrote them — no brand bias, just the argument.
- Divergence — with a score. A 0–100 read of how far apart the council landed, how many camps formed, and who the outlier is. It measures agreement, not correctness.
- Debate (optional). One bounded rebuttal round: each advisor revises or holds, and says why. Who moved, who held.
- Synthesis & Dissent. A decision-ready distillation — plus the outlier's full answer spotlighted, because the majority can be confidently wrong together.
- You decide — and log it. Record your decision in the journal, and later, how it actually turned out.
Already have an answer from ChatGPT, Gemini, or anything else? Paste it in as a guest seat. It takes a place at the table as one more anonymous advisor — the council critiques it blind, without knowing it came from outside, and it counts in the divergence score, the dissent, and the synthesis. One connected advisor plus a guest is already a council of two.
- 🧠 Twelve backends — Claude · GPT (OpenAI) · Gemini · DeepSeek · Grok (xAI) · Mistral · Perplexity · OpenRouter · Ollama (local, needs Ollama running) · Apple Intelligence (on-device, free — needs macOS 26 on Apple-silicon) · two custom OpenAI-compatible endpoints (llama.cpp, LM Studio, vLLM, a second Ollama box — with a test-connection button that pulls the server's real model list)
- ⚡ Live streaming answers, side by side
- 🎭 Distinct personas per seat (Analyst · Practitioner · Skeptic) for real divergence — not three ways of saying the same thing
- 🎭 Guest seat — paste an answer from any other AI; it's blind-reviewed as one more anonymous advisor
- ⚖️ Neutral chair (optional) — a model that moderates instead of answering: it writes the divergence and synthesis, and closes a debate with a neutral summary
- 😈 Devil's Advocate role to pressure-test the consensus
- 📊 Divergence score — 0–100 how far apart the council landed, camps, and the outlier; agreement, not correctness
- 🗣️ Bounded debate — one optional rebuttal round; original answers stay tucked underneath so you see what moved
- ☑️ Selective deliberation — after reading the answers, choose which of them go into peer review
- ❗ Dissent — the outlier's full answer, surfaced on its own to judge for yourself
- 📓 Decision journal — log what you chose, set a reminder, and macOS notifies you when it's time to record how it turned out (local only, no account)
- 🧭 Two layouts — Flow (one page; analysis appears beneath the answers) or Classic (each stage its own screen)
- 🖼️ Vision — drop in an image for the models that support it
- 📎 Attach a document — add a
.md/.txtto your question (or drag it in) and the whole council reads it before weighing in; the file stays out of your saved history - 💸 Calm cost estimate — a running token/$ tally, your spend at a glance, and an optional spend alert
- 📤 Export — Markdown, PDF, image, or a paste-ready decision memo; share councils as importable presets
- ⌨️
councilCLI — the same engine in your terminal: pipe documents in, get JSON out, gate CI on divergence - 🔄 In-app updates — new versions install from inside the app (it's an unsigned build, so macOS won't hand the new version the old one's Keychain items — Council asks you to re-enter your keys once, inline)
- 🔗 Engram bridge (optional) — if you already run Engram, a journal decision can be written into your local Engram store, so your other AI tools can recall it
- 💾 Local history of every session (CLI runs land there too)
- 🪟 Native SwiftUI — real Liquid Glass on macOS 26, a graceful material fallback on 14+
Council is 100% local. No account, no server, no telemetry.
- Your API keys live only in the macOS Keychain. They're masked in the UI and are never written to disk, exports, logs, or session files.
- Each key is sent only to that provider's own endpoint, over HTTPS. Local backends — Ollama and your own OpenAI-compatible servers — go wherever you point them, which is
localhostunless you change it in Settings. - You pay the providers directly with your own keys — Council never sits in the middle.
Don't have a key yet? Council links you straight to each provider's console from the key-entry step.
Homebrew — one command:
brew install --cask albertofettucini/council/councilOr download the macOS build from the latest release, unzip it, and drag Council.app to Applications. Requires macOS 14 or later.
⚠️ First launch: Council isn't signed with a paid Apple certificate (it's a free, solo, open-source project), so macOS blocks it the first time. OpenCouncil.apponce; when macOS says it "could not verify" the app, click Done. Then go to System Settings → Privacy & Security, scroll down to Security, click Open Anyway next to the Council message, and confirm with your password or Touch ID. Do it right away — that button only shows for about an hour after the blocked launch. It opens normally after that. (Still on macOS 14? Right-clickCouncil.app→ Open works there too.)Want to skip that step?
brew install --cask --no-quarantine albertofettucini/council/council, or after a manual download:xattr -dr com.apple.quarantine /Applications/Council.app.Rather build it yourself? The whole app is in this repo — see Build from source.
git clone https://github.com/albertofettucini/Council.git
cd Council
open Council.xcodeproj # Xcode 26 or later (the Liquid Glass path needs the macOS 26 SDK; the built app still runs on macOS 14+)
# ⌘R to runThree SPM dependencies, all resolved automatically by Xcode: Sparkle (in-app updates), CouncilKit (the engine — a separate package at github.com/albertofettucini/CouncilKit, shared with the CLI), and Engram (only for the optional decision-memory bridge). Otherwise pure SwiftUI + Foundation.
The same engine, in your terminal — for scripting, CI, and piping documents in:
git clone https://github.com/albertofettucini/CouncilKit.git
cd CouncilKit && swift build -c release
cp .build/release/council /usr/local/bin/ # or anywhere on your PATH
council keys set claude # keys go to the macOS Keychain (shared with the app)
council "should we ship now or wait?" --seats claude,gpt,gemini
council "review this" --file design.md --md # attach a document, get a decision memo
cat design.md | council "review this" # …or pipe it in on stdin
council "..." --guest chatgpt-answer.txt # seat an outside answer in the blind review
council "..." --chair claude # a neutral chair moderates instead of answering
council "..." --json # structured output (schema council.cli.v1)
council "..." --fail-above 40 # CI gate: exit 1 if the council diverges too muchCLI runs land in the app's history, so you can reopen them in the UI. council --help has the full flag list.
"Should a two-person startup adopt microservices on day one?"
Claude weighs the trade-offs, GPT pushes back on premature complexity, Gemini brings the ops angle. The peer-review round catches where one of them overreached, Divergence shows the real fault line, and Synthesis hands you a decision-ready summary.
- More local backends out of the box
- Recall past decisions when a new question resembles an old one
Issues and PRs welcome. Council is one person's project — keep changes focused and the privacy guarantees intact: no telemetry, and keys never leave the Keychain.
Questions, ideas, or feedback? Open an issue — or reach me at joseph.thecouncil@gmail.com.
Engram — one shared, local memory for all your AIs. Import your ChatGPT/Claude history once; every MCP client can recall it.
MIT © 2026 Joseph
Made for people who'd rather weigh a few good opinions than trust one.