New: a Local engine. Run Reply Better AI against your own model with Ollama, LM Studio, or any OpenAI-compatible server, with no key, nothing leaving your machine, and offline support (works on Firefox too). Plus settings improvements: the popup now opens on your saved default writing style; the Model section shows which model each engine uses (the OpenRouter picker for OpenRouter and Auto, a read-only summary for on-device, Groq, and local); settings shows only the key the active engine needs; the navigation works as real tabs; and the inline button no longer hides behind Grammarly's icon.

CORS note for the Local engine: LM Studio needs its CORS toggle turned on; on Firefox, Ollama needs OLLAMA_ORIGINS set to allow the extension. The setup guide covers both.

New in 1.5.0:

• Reply mode. One inline button that becomes a Reply button when you select the messages you are replying to, or an Improve button when you have a draft. Pick a tone (Match thread, Professional, Friendly, Concise, or Summarize) or type your own instruction in any language, and the reply comes back in that language.
• New free engine. Add your own free Groq key for fast generation, alongside OpenRouter's 500+ models. Choose your engine in settings, or leave it on Auto to use the best one you have set up and fall back automatically if one is busy. The active engine is always shown.
• Auto, Fastest free. One pick that routes to the fastest available free model and fails over automatically when one is busy.
• A word-level Changes diff, Regenerate for another take, and one-tap Insert with Undo.
• Redesigned interface with dark mode and a standalone full-tab settings page.
• Clearer, provider-neutral error messages and more robust streaming.

What's new in 1.3.0

• Fresh new look with a cleaner design and automatic dark mode.
• Rewrites now stream in live, so you watch the result appear as it's written.
• New Changes view highlights exactly what was edited, word by word.
• Regenerate cycles through alternative versions until one fits.
• The inline ✍️ button now opens a review panel: pick a style, preview the result with its changes, then insert it. Prefer one click? Switch it to instant rewrite in settings.
• Redesigned model picker and a rebuilt full-page settings screen.

v1.2.3 — Bug fix: the "model is no longer available" banner used to keep reappearing every time the popup opened, even after switching to a working model. Picking a new model now dismisses the notice automatically.

Major update — v1.2.x.

• Dynamic model picker: search and pick from 500+ AI models on OpenRouter (Claude, GPT, Gemini, DeepSeek, Llama and more) with live pricing and a free-only filter. The previous hardcoded three-model dropdown is gone.

• Smarter inline button: the ✍️ button now only appears on long-form message composers (Gmail, Twitter/X, LinkedIn, Slack, Discord, etc.) instead of popping up on every search bar and form field.

• Friendlier errors: clear messages for invalid API keys, unavailable models, rate limits, and offline state — no more generic alert popups.

• New full-tab settings page in addition to the toolbar popup.

• API key is now stored in local storage instead of cloud-synced storage, so credentials no longer roam between your browser profiles. Existing keys are migrated automatically on first run; nothing for you to do.

• Privacy policy: https://github.com/dantnan/reply-better-ai/blob/main/docs/privacy.md

Version 1.1.0 - AI Model Update

🚀 What's New:
- Upgraded to latest and most powerful free AI models:
• DeepSeek V3 - Best overall performance with 131K context window
• Gemini Pro - Google's advanced model with large context support
• DeepSeek R1 - Open source model with excellent reasoning capabilities

🔧 Improvements:
- Better text improvement quality with more advanced models
- Enhanced instruction following for custom prompts
- Improved performance and reliability
- Clearer model descriptions in the dropdown menu