Conversational Reviews
ThinkReview isn’t just a static report. You can actually “talk” to your review results to ask follow-up questions, request clarifications, or dive deeper into specific suggestions.
How it works
Section titled “How it works”Once a review is generated, a chat interface becomes available. When you send a message:
- Context Awareness: The system sends your current code patch, the original review summary, and your previous conversation history to the AI.
- Context Limits: To maintain speed and accuracy, we send the last 6 to 11 messages of your conversation (depending on the model used).
- Smart Truncation: If your code patch is very large, our smart truncation engine ensures the AI still sees the most important parts while answering your questions.
What you can ask
Section titled “What you can ask”- Clarifications: “Can you explain why you suggested using async/await here?”
- Alternatives: “What’s another way to solve this security issue without changing the library?”
- Deep Dives: “Show me an example of how to implement this best practice.”
- Confirmation: “If I change line 45 to X, will that fix the vulnerability?”
Key Features
Section titled “Key Features”Suggested Questions
Section titled “Suggested Questions”Every review ends with 3 suggested follow-up questions. These are specifically generated based on your code to help you explore the feedback. You can click these to start a conversation instantly.
Multi-Model Support
Section titled “Multi-Model Support”Conversational reviews use the best model available for your subscription tier. If one model fails to respond, the system automatically tries a fallback model to keep the conversation going.
Language Support
Section titled “Language Support”You can ask questions in your preferred language, and the AI will respond accordingly, maintaining the context of the technical discussion.
Best Practices
Section titled “Best Practices”- Be Specific: Mention file names or line numbers when asking about specific parts of the code.
- One Question at a Time: You’ll get better results by asking focused questions rather than complex multi-part queries.
- Stay in Scope: The AI’s context is focused on the current patch and MR. Asking about unrelated parts of the codebase may result in generic answers.
TL;DR: Use the chat interface to ask questions about your review. It remembers your conversation context and uses your code patch to provide specific, actionable answers.