Quorum — Multi-LLM Consensus for VS Code
Quorum runs a single prompt across 8+ frontier LLMs (Claude, GPT, Gemini, Llama, DeepSeek, Mistral, Qwen, Phi) directly inside VS Code, scores the answers for semantic agreement, and surfaces exactly where the models disagree — because disagreement between strong models is the most valuable signal you can get before shipping code, taking a security call, or making a regulatory commitment. Backed by the hosted Quorum consensus engine; BYOK supported so your provider keys never leave your machine.
Install
- Download
quorum-vscode-0.1.0.vsix from the Releases page.
- In VS Code: open the Extensions sidebar, click the
… menu, choose Install from VSIX…, and pick the .vsix file.
- Open Settings (
Cmd+, / Ctrl+,), search quorum, and set at minimum quorum.endpoint and either quorum.apiKey or your BYOK provider keys.
Or build from source:
git clone https://github.com/jaquelinejaque/sovereignchain.git
cd sovereignchain/quorum-vscode
npm install
npm run build
npm run package
code --install-extension quorum-vscode-0.1.0.vsix
Commands
| Command |
What it does |
Quorum: Ask |
Free-form prompt, returns a consensus answer in a side panel |
Quorum: Ask about selection |
Combines your selection with a follow-up question |
Quorum: Explain selected code |
Multi-model explanation of the highlighted code |
Quorum: Review selected code (bugs/security) |
Adversarial code review across all configured models |
Quorum: Compare two implementations |
Pick two snippets, get a structured side-by-side comparison |
Quorum: Open settings |
Jumps straight to the Quorum settings page |
The Explain and Review commands also appear in the editor right-click menu whenever you have a selection.
Settings
| Setting |
Default |
Description |
quorum.endpoint |
https://quorum-ai.dev |
Quorum server endpoint |
quorum.apiKey |
"" |
X-Quorum-API-Key header (leave empty to use BYOK mode) |
quorum.providers |
["gemini-flash","llama-3.3-70b","deepseek-v3","claude-sonnet-4-6"] |
Models to query in parallel |
quorum.maxLatencyMs |
30000 |
Per-query timeout in milliseconds |
quorum.showCostInline |
true |
Show per-query cost inline in the result panel |
Sample workflow
- Highlight a function in any editor (TypeScript, Python, Rust, anything).
- Right-click → Quorum: Review selected code (bugs/security).
- Quorum fans the snippet out to all configured models, scores their agreement, and renders a panel showing the consensus verdict, the dissenting opinions, and a per-model latency/cost breakdown.
The same pattern works for explanations (Explain selected code), free-form questions about a snippet (Ask about selection), and side-by-side comparisons (Compare two implementations).
Screenshots
Webview screenshots will be added in v0.2.0 once the consensus result panel ships. The phase-1 release uses VS Code notification toasts to confirm command wiring.
Pricing
Free during beta. Pro £49/mo when GA, which covers hosted consensus, all upstream provider calls, and priority routing. BYOK is fully supported — point quorum.endpoint at your own deployment or leave quorum.apiKey empty and supply provider keys directly; in BYOK mode you only pay your upstream providers.
Known limitations
Being honest about what 0.1.0 is and isn't:
- Depends on the hosted Quorum engine. You need either a
quorum-ai.dev API key or your own self-hosted Quorum instance for the commands to return real answers. Without one, the commands fire but receive no consensus.
- No streaming yet. Results arrive as a single payload once the slowest configured model finishes (bounded by
quorum.maxLatencyMs). Streaming lands in v0.2.0.
- No inline diagnostics yet. Findings render in a side panel, not as squiggles in the editor. Inline diagnostics + Code Lens land in v0.3.0.
- Phase-1 commands are stubs. This release ships the full command/menu/settings surface and toasts a confirmation; the real consensus calls and webview land in v0.2.0 (tracked in
CHANGELOG.md).
Links
