ThinkReview - Version 1.5.1 Release Notes
Release Date: 15 February 2026
---
🎯 What's New
Ollama Configuration Improvements 🖥️
More control over local AI behaviour: When using Local Ollama, you can now tune generation options directly in the extension popup. These settings apply to both the initial code review and follow-up conversation.
- Temperature (min 0, max 2): Controls randomness; lower values give more deterministic, focused reviews. Default: 0.3
- Top P (min 0, max 1): Nucleus sampling; helps control diversity of responses. Default: 0.4
- Top K (min 1, max 200): Limits how many top tokens are considered. Default: 90
All three fields appear in the Ollama configuration section (under AI Provider when Local Ollama is selected). Values are validated and clamped to the allowed ranges when you save, so invalid or out-of-range entries are corrected automatically. The same limits are enforced when sending requests to Ollama, so your settings are always safe for the API.
Under the hood: Clamping logic is centralized in a shared utility (utils/ollama-options.js) used by both the popup and the Ollama service, so limits and defaults stay consistent everywhere.
---
Documentation Updates 📖
- Recommended model: The Ollama Setup Guide now recommends Codestral for the best code review results (Mistral AI’s 22B code model, 32k context, 80+ languages).
- Alternative model: gpt-oss (OpenAI’s open-weight models; 20B/120B, 128K context) is documented as a supported option.
- The main README’s “Local AI with Ollama” section now points to Codestral as the recommended model and gpt-oss as an alternative.
---
🐛 Bug Fixes & Improvements
- Conversation history on GitHub and Azure DevOps (and Bitbucket): Fixed a bug where conversation history from a previous PR could appear when opening a new PR without refreshing the page. On these SPA-based sites, the extension now clears in-memory conversation history and the chat log UI whenever a new review is started (including when the panel is re-injected after SPA navigation), so each PR shows only its own review and follow-up messages.
- Ollama options: Temperature, Top P, and Top K are now persisted with your Ollama URL and model; they are loaded when you open settings and applied to every Ollama request (review and conversational).
- Consistent limits: Min/max limits are shown in the popup labels and enforced on save, load, and in the service, so you never send invalid values to Ollama.
---
📝 Usage Tips
1. Tuning Ollama
- In the popup: AI Provider → Local Ollama → set Temperature, Top P, and Top K as needed, then Save Settings.
- Lower temperature (e.g. 0.2–0.4) usually gives more consistent, focused reviews; higher values can increase variety.
2. Best results with Ollama
- We recommend Codestral (ollama pull codestral) for code review quality. See the Ollama Setup Guide for details and other options like gpt-oss.
---
📞 Support
Having issues or suggestions? We'd love to hear from you:
- Bug Reports: Click the "Report a 🐞" button in the review panel
- Feedback Form: Share Your Feedback
- Rate Us: Leave a review on the Chrome Web Store
---
Thank you for using ThinkReview! 🚀
Ollama users: tune Temperature, Top P, and Top K from the popup and try Codestral for best results.