Quorum
Compare multiple AI models at once. Quorum runs a panel — a council — of AI models that deliberate, disagree, and converge on one recommendation instead of a single model's guess: a built-in second opinion for code review, debugging, architecture, design, and more. Copilot-first.

Features
- 7 council types — Spec Review, Debugging, Code Review, Design Review, Research Synthesis, Architecture Tradeoff, PM Strategy
- Multi-model panels — 3–5 models deliberate, disagree, and converge on a recommendation
- Decision memos — Structured output with recommendation, dissent, confidence, and evidence
- Council Threads — Continue the conversation with follow-up questions in an editor tab
- Code context (opt-in) — Ground councils on your repository: selected files and their direct imports are captured as a redacted, content-hashed snapshot and cited in the memo
- Auto-routing — Describe your problem and let Quorum pick the right council type
- Copilot-first — Works with your existing GitHub Copilot subscription out of the box
Quick Start
1. Install
Search for Quorum in the VS Code Extensions view, or install from the Marketplace.
2. First Run
- Open the Quorum sidebar (activity bar icon)
- Type your question or paste a spec/code snippet into the composer
- Select a council type (or let auto-routing choose)
- Click Run Council — models deliberate and produce a decision memo

3. Follow Up
Once a memo is produced, open the Council Thread to ask follow-up questions. The thread maintains full context from the original deliberation.

Council Types
| Council |
Purpose |
| Spec Review |
Find contradictions, missing requirements, ambiguity |
| Debugging |
Compare hypotheses, map evidence, eliminate weak explanations |
| Code Review |
Multi-perspective review with explicit disagreement |
| Design Review |
Evaluate design decisions and tradeoffs |
| Research Synthesis |
Identify flaws, blind spots, and missing evidence |
| Architecture Tradeoff |
Structured tradeoff analysis with weighted criteria |
| PM Strategy |
Product strategy evaluation and prioritization |
Configuration
| Setting |
Default |
Description |
routerCouncil.provider.mode |
copilot-only |
Provider strategy: copilot-only, copilot-preferred, or mixed |
routerCouncil.models.defaultPanelSize |
3 |
Number of models per council run (1–5) |
routerCouncil.councils.maxExtraIterations |
2 |
Extra critique rounds when disagreement is high (0–2) |
Requirements
- VS Code 1.100+
- GitHub Copilot subscription (for
copilot-only and copilot-preferred modes)
- For
mixed mode: an OpenAI-compatible API endpoint (see Developer Guide)
Development
See the Developer Guide for provider setup, architecture notes, troubleshooting, and test commands.
Naming & research
Quorum was previously named Router Council. That earlier name is intentionally retained in some internal identifiers for backward compatibility — command and setting IDs (routerCouncil.*), the on-disk router-council run-storage folder, and prompt-template IDs — so existing keybindings, settings, and saved runs keep working. All end-user-facing text is "Quorum".
The extension is also the software artifact studied in the research paper "When Does Iteration Pay? Characterizing Dissent-Driven Deliberation in Multi-Model AI Councils", which refers to the tool by its development name Router Council for citation stability. The paper and the extension are distinct publications.
License
See LICENSE.