Major framework modernization with Vinext and Vite 8 (rolldown-based). Replaced LangChain with Vercel AI SDK across the chat pipeline. Improved error handling in model loading and database operations. Fixed Worker deployments, CommonJS/ESM compatibility, and frozen lockfile issues.
Model Update: Changed default model for OpenRouter provider from Kimi 2.5 to Nemotron 3 Super 120B for all users and guests. Updated chat configuration to prioritize Nemotron models across the platform.
Authentication Fix: Resolved "Invalid Base64 character" error in Better Auth by implementing proper BETTER_AUTH_SECRET configuration for Cloudflare Workers. Added automated setup script (setup-auth-secret.sh) and comprehensive documentation for secret management.
UI/UX overhaul with migration from @opennextjs/cloudflare to Vinext. Implemented macOS-style category dock with theme switching. Consolidated authentication with better-auth 1.6.14 and Web Crypto API. Added responsive layouts, dynamic island TOC positioning, and font controls. Fixed vite-rolldown aliasing and turbopack build failures. Enhanced deployment scripts.
Editor and authentication enhancements. Integrated Google One Tap with FedCM and incognito mode. Added Shiki code highlighting, Mermaid diagrams, word count modals, and document export. Implemented file management with lazy initialization. Expanded settings with API key controls and sign-out. Added reason-editor module with new plugins. Refactored database schema for cross-environment compatibility. Updated OpenNext Cloudflare deployment with PWA assets.
Major V2 rewrite with fundamental restructuring. Optimized project structure and removed deprecated dependencies. Reorganized scraper infrastructure with rebuilt Next.js configuration. Implemented XGBoost statistical analysis, replacing neural utilities. Overhauled documentation and README. Refactored research agent components and migrated chat/article modules. Enhanced editor with font customization and menu improvements. Improved Cloudflare Workers configuration.
Beta V1 major release with complete feature set. Comprehensive login and user management. Integrated editor with full capabilities. Docusaurus documentation with OpenAPI and TypeDoc support. Automatic API reference generation. First production-ready version.
Search infrastructure improvements. Enhanced Docker-based search system with better reliability. Fixed YouTube integration for video content. Added DOCX file format support. Fixed USearch vector accuracy issues. Enhanced content extraction capabilities.
Topic modeling and citations. Completed SeekTopic integration for topic extraction and analysis. Standardized citation formatting platform-wide. Improved README with better examples. Focused on academic and research features.
Core algorithm implementations. Built VSEARCH (Vector Similarity Embedding Approximation) as custom vector search. Added category systems for organization. Introduced Tardigrade web crawler for distributed crawling. Expanded documentation with categories. Enhanced main UI. Established algorithmic foundations.
Content extraction and NLP. Ported Trafilatura.js from Python (33 files) for article extraction. Enhanced Readability2 accuracy. Added UMAP dimensionality reduction. Implemented HNSW vector search with demos. Modularized extractors (Readability, Postlight). Adopted "code as art" philosophy. Implemented TypeDoc documentation. Added YouTube embed API with transcript optimization. Enhanced extension with CORS support.
Search algorithms and autocomplete. Implemented DSEEK keyphrase extraction with TextRank, WikiIDF, and noun edge-grams. Added query autocomplete with live demo. Introduced new compression formats. Integrated OpenEnglishWordnet and 35k Wikipedia pages. Added RAG use case. Implemented Wiki BM25 with 1M/2M datasets. Published results demo. Enhanced search quality and linguistic capabilities.