AI Orchestra

Ask one question to multiple AIs. They debate each other — you get one unified answer.

---

What it does

AI Orchestra is a macOS desktop app. You type one question. It sends that question to several AI models at the same time (Claude, ChatGPT, Gemini, and any OpenAI-compatible model you add). The models then review each other's answers, and a final model merges everything into one clear answer.

Features

• Orchestra mode — models answer, critique each other for 1–5 rounds, then a judge writes one final answer.
• Split mode — see each model's answer side by side, no debate.
• Many providers — Claude, ChatGPT, Gemini built in. Add any OpenAI-compatible API (xAI, DeepSeek, OpenRouter, local Ollama, …) with just a base URL.
• Model picker — choose models from a dropdown (cheap → expensive), or type a model id by hand.
• Token usage — see input/output tokens per model and per run.
• History — every conversation is saved, auto-named, and searchable. Continue old chats (each chat keeps its own memory).
• Two languages — English and Turkish. Follows your system language, remembers your choice.
• Daily model refresh — the live model list updates once a day (and on demand).
• Private by design — API keys are stored encrypted in the macOS Keychain. Nothing is sent anywhere except the AI providers you choose.

Screenshots

Left: each model answers, then critiques the others. Right: the judge merges everything into one answer, with per-model token usage.

What's new (v1.3.0)

• Streaming — answers appear as they're written (split columns and the final unified answer).
• Cost estimate — approximate USD per model and total, from real token usage.
• Retry a single model (↻) and Regenerate the unified answer.
• Reorder participants; opening a past chat restores its mode and rounds.

Earlier (v1.2.2): security hardening (keys never stored unencrypted, sent only to https/localhost, navigation guard), failed-synthesis/stop keeps answers, Cancel/Esc revert edits.

• Error logging to ~/.ai-orchestra/error.log (open from Help → Open Logs Folder).
• Fixed the "app is damaged" message some users saw after downloading.
• App version and date shown in the corner.

See the full CHANGELOG.

How it works

Your question
     │
     ├──► Claude  ─┐
     ├──► ChatGPT ─┤  1) all answer in parallel
     └──► Gemini  ─┘
                   │
                   ▼  2) each model reads the others and improves its answer (N rounds)
                   │
                   ▼  3) a judge model merges everything
                   │
              One unified answer

Download

Grab the latest .dmg from the Releases page, open it, and drag AI Orchestra into Applications.

First launch on macOS. The app is signed but not notarized (no paid Apple Developer account yet), so macOS blocks it the first time. If you see "AI Orchestra is damaged" or "unidentified developer", run this once in Terminal — it clears the download flag:

xattr -dr com.apple.quarantine "/Applications/AI Orchestra.app"

Then open the app normally. (You can also try right-click → Open, but the Terminal command always works.) Or just run it from source (below) to skip this entirely.

Getting started

You need API keys for the providers you want to use (Anthropic, OpenAI, Google AI Studio).

git clone https://github.com/ahmetrende/ai-orchestra.git
cd ai-orchestra
npm install
npm start

On first launch, open Settings (⚙) and paste your API keys. Then type a question and hit Send.

Build a macOS app

npm run dist        # creates a .dmg in dist/

Or build an unsigned .app for local use:

npx electron-builder --mac --arm64

Where your data lives

Everything stays on your Mac, in ~/.ai-orchestra/:

• config.json — settings and API keys (keys are Keychain-encrypted)
• history.json — your conversations
• models-cache.json — the cached model list

This folder is kept even if you delete and reinstall the app.

Tech

Electron · vanilla JS (no build step) · marked + highlight.js for Markdown/code rendering.

Credits

Built with these open-source projects (bundled in src/vendor/):

• marked — Markdown parser (MIT)
• highlight.js — code syntax highlighting (BSD-3-Clause)
• DOMPurify — HTML sanitizer (Apache-2.0 / MPL-2.0)
• @lobehub/icons — AI provider brand logos (MIT)
• Electron — desktop runtime (MIT)

Provider names and logos (OpenAI, Claude, Google Gemini, …) are trademarks of their respective owners and are used only to indicate compatibility.

License

MIT

---

Made by Ahmet Rende · GitHub