Everything shipped in Mindloop, version by version.
v0.4.0
2026-08-07
New
Humanizer: rewrites stiff or AI-sounding text to read naturally, with 4 tone options (Natural, Casual, Professional, Academic) — powered by Yuu no Sekai
20 more color themes (Kyoto, Nordic, Desert, Midnight, Coral Reef, Lavender Fields, Autumn, Aurora, Rose Gold, Emerald, Cyberpunk, Vintage Paper, Ocean Depths, Volcanic, Mint, Peach, Slate, Amethyst, Honeycomb, Galaxy — 25 total now), each with a generated, tileable PNG background pattern
Technical
Theme backgrounds are procedurally generated (SVG patterns rasterized to PNG at build time via sharp) rather than stock art, keeping the whole set under 250KB
v0.3.0
2026-08-07
New
5 color themes (Mindloop, Tokyo Night, Sakura, Forest, Sunset), each with a light and dark variant, picked from Settings or a quick swatch dropdown in the sidebar
Explicit theme picker and Log out button in the sidebar
Connected AI_PROVIDER="goat-ai": a self-hosted, OpenAI-compatible model gateway (api.gariyuuu.com) serving a custom model ("Yuu no Sekai"), as an alternative to Anthropic/OpenAI/Gemini
Technical
The self-hosted gateway's model defaults to a verbose "thinking" mode that burns ~10x the output tokens — disabled via a fetch middleware that patches the outgoing request body, since the AI SDK's typed provider options don't expose that field for OpenAI-compatible backends
v0.2.0
2026-08-06
New
Semantic (vector) search for the AI Tutor's retrieval, via pgvector on Neon — falls back to keyword search automatically when no embedding-capable key is configured
LaTeX formula and syntax-highlighted code rendering across flashcards, summaries, and the AI Tutor
AI-generated short definitions for Match mode, folded into the main generation call at no extra cost
Retry banner on failed study-set generation with a human-readable reason and a working retry action
Verified, live Clerk webhook syncing user create/update/delete events to the database
Improved
Match mode now shows meaning-preserving short definitions instead of naive truncation when available
Fixed
A real concurrency bug in demo-content seeding: Next.js's automatic link prefetching could fire multiple concurrent requests for a brand-new user, and the request that lost the seeding race used to render immediately with no data instead of waiting for the winner to finish
AI Tutor chat could hang on a permanent loading spinner when the upstream AI call failed after the response had already started streaming
Theme selector in Settings could cause a hydration mismatch on first load
Technical
Neon HTTP driver queries now explicitly bypass Next.js's fetch cache (no-store) to prevent stale reads immediately after a write in the same request
Sidebar navigation links no longer prefetch, removing redundant background requests on every page view
v0.1.0
2026-08-06
New
Study guide importing via paste-text or document upload (PDF, DOCX, TXT, Markdown)
AI-powered flashcard generation with contextual, source-grounded examples
Flashcards, Learn, Test, Match, Blast, and Active Recall study modes
AI Tutor grounded in your own study material, with quiz/explain/socratic/exam-prep modes