CouncilAI routes your questions to Claude, GPT, Grok, or any of the hundreds of models available through OpenRouter — using your own API keys. Compare answers, run a full deliberation, or let a sequential pipeline draft, critique, and refine.
Get CouncilAI for WindowsWindows 10/11 · 64-bit · Bring your own API keys · One-time app purchase
How it works
Add API keys for the providers you already use or want to try. CouncilAI never generates anything on its own — it routes, compares, and refines using the models you've connected.
Add an API key for Claude, GPT, Grok, and/or OpenRouter in Settings. Each key stays on your device and talks directly to that provider — never through a CouncilAI server.
Ask a single model directly, run Council Mode for a full multi-model deliberation, Compare Mode to see answers side by side, or Sequential Mode for a draft → critique → refine pipeline.
The See My Thinking panel shows which model answered, why, and — in Council Mode — the full scoring breakdown between every model that responded.
Features
Every mode is designed around one idea: multiple models, working together, with nothing hidden.
Every connected model answers in parallel. A scoring pass and an AI-judged comparison pick the strongest response — you see every answer, not just the winner.
One model drafts a response, a second critiques it, a third refines it into a final answer — genuinely different models catching each other's blind spots.
Pick any two connected models and see their raw answers side by side — word count, length, and an agreement indicator show you where they converge or diverge.
A transparency panel showing exactly which model answered, why, and how it scored against the others — no black box.
After a Council or Sequential round, CouncilAI automatically continues the conversation with the winning model — no need to re-run the full council for every follow-up.
An optional running token and cost counter shows exactly what your session is costing, since every message uses your own API billing.
How pricing works
CouncilAI itself is a one-time purchase for the app. Every message you send is billed by whichever provider answered it — the same rate you'd pay calling that provider's API directly.
Sign up with Anthropic, OpenAI, xAI, or OpenRouter directly. CouncilAI walks you through it if you've never set up an API key before.
Most conversations cost fractions of a cent to a few cents. You're billed by the provider, at their standard rate.
OpenRouter offers genuinely free models with rate limits — a way to try CouncilAI's multi-model modes at zero cost.
An optional in-app counter shows running token usage and estimated cost for your current session.
Privacy
CouncilAI doesn't operate any backend. Your messages go directly from your device to whichever provider you've connected — the same as if you called their API yourself.
API keys are saved locally using your OS's secure credential store — never sent anywhere except directly to that provider.
There's no sign-up, no login, no CouncilAI-hosted account of any kind.
We don't operate a backend that sees your conversations. Requests go straight from your device to the provider you chose.
Once a message reaches Claude, GPT, Grok, or an OpenRouter-hosted model, that provider's own privacy policy governs how it's handled — the same as using their product directly.
Get started
Download the installer, connect at least one provider, and start your first council.
Get CouncilAIWindows 10/11 64-bit · Requires at least one provider API key · Provider usage billed separately by that provider
Changelog
🔌 Rebuilt around Claude, GPT, Grok & OpenRouter
CouncilAI no longer bundles or runs local models. Connect your own API keys and route between real frontier models — plus 500+ more through OpenRouter.
🔀 Auto-continue after Council or Sequential rounds
Follow-up questions no longer re-run the entire council — CouncilAI continues with the winning model automatically.
📊 Live token & cost tracking
An optional running counter shows session tokens and estimated cost, since usage is now billed by your connected providers.
✨ Real-time streaming in Sequential Mode
Draft, critique, and refine stages now stream live instead of appearing all at once.
Early beta releases ran entirely on locally-bundled models (Llama, Gemma2, Phi-3, Mistral) via a bundled Ollama runtime. This local-only mode has been retired in favor of the provider-based architecture in v2.0, which gives access to genuinely frontier-class models.