# Opinions

Opinions lets you query multiple AI models at once with a question or implementation plan. Each model analyzes your codebase independently, then a synthesis pass converges their responses into a single consensus — so you get a well-rounded answer, not just one perspective.

<!-- ![Opinions showing synthesis and model tabs](/images/docs/TODO.png) -->

## How It Works

1. **You ask a question** — This could be an architecture question, implementation plan, code review request, or any technical question about your codebase.
2. **Multiple models respond in parallel** — Each selected model analyzes your codebase independently in read-only mode.
3. **Synthesis** — Once all models have responded, a final pass distills their answers into a converged recommendation, highlighting where they agree and where they differ.

## Starting an Opinions Run

Open the Opinions dialog from the workspace toolbar or use the command palette (**&#8984;K**).

<!-- ![Opinions dialog with model selection](/images/docs/TODO.png) -->

Select two or more models to query, then type your question. All models receive the same prompt and have read-only access to your codebase — they can explore files and search code, but cannot make any changes.

## Reading the Results

When the run completes, the results appear in the activity feed.

<!-- ![Opinions results](/images/docs/TODO.png) -->

### Synthesis

The **synthesis** section is shown first. It summarizes the consensus across all models, noting:

- **Points of agreement** — Where all models recommend the same approach
- **Points of divergence** — Where models disagree, with reasoning from each side
- **Final recommendation** — A converged answer that weighs the different perspectives

### Individual Responses

Click the model tabs to read each model's full response. If you selected the same model more than once, the tabs are numbered (e.g., "#1", "#2") so you can distinguish between them.

Each response includes a duration indicator showing how long that model took.

## Use Cases

- **Architecture decisions** — "Should we use WebSockets or SSE for real-time updates?" Get perspectives from multiple models before committing to an approach.
- **Code review** — "Review this authentication flow for security issues." Multiple models catch different classes of issues.
- **Implementation planning** — "How should we add caching to the API layer?" Compare approaches before writing code.
- **Debugging** — "Why might this test be flaking?" Different models may identify different root causes.

## Tips

- **Use Opinions before starting complex work.** Getting consensus upfront avoids going down a suboptimal path.
- **Ask specific questions.** "How should we handle pagination in the search API?" works better than "How should we improve the API?"
- **Select diverse models** when possible. Different models have different strengths and are more likely to surface complementary insights.