Today's trending projects reflect two tensions playing out across the AI landscape at once: a drive to make models genuinely useful in narrow, high-stakes domains rather than just generally capable, and a growing awareness that AI-generated text has a recognizable voice — one many writers and developers are now actively trying to suppress. Both problems are getting serious engineering attention.
Kronos: An Open-Source Foundation Model for Financial Candlesticks
shiyu-coder/Kronos reached 27,000 stars today as the first open-source foundation model built specifically for financial candlestick (K-line) data. Trained on OHLCV records from over 45 global exchanges, it uses a two-stage approach: a specialized tokenizer converts continuous market data into hierarchical discrete tokens, then a decoder-only Transformer learns to model those tokens autoregressively. Model sizes run from 4.1M to 499.2M parameters, and the framework supports price forecasting across multiple timeframes, batch predictions for portfolios of assets, and fine-tuning for specific markets or strategies. Raw predictions still need portfolio optimization layered on top for production use, but the project is a meaningful step toward domain-adapted models that treat financial data as the high-noise, structure-rich signal it is rather than generic sequences.
Stop-Slop: Score Your Prose, Rewrite What Sounds Like a Robot
drm-collab/stop-slop gathered over 6,000 stars with a pointed premise: AI-generated prose is identifiable, and that's a problem worth solving with a tool rather than just willpower. The Claude Code skill evaluates submitted text across five dimensions — directness, rhythm, trust, authenticity, and density — and automatically rewrites passages that score below 35 out of 50. The rewrite targets eight specific failure modes: filler language, repetitive sentence structures, passive constructions, vague abstractions, absent imagery, uniform pacing, condescending hedges, and the kind of tidy-but-hollow phrases that tend to close AI-written paragraphs. It re-scores after rewriting to confirm the improvements are real. For anyone publishing AI-assisted content where voice and credibility matter, this addresses a gap that grammar checkers and style linters were never designed to fill.
github.com/drm-collab/stop-slop
Nango: One Platform for 800+ API Integrations, AI-First
NangoHQ/nango added 442 stars today, continuing steady growth to over 9,300 total, and its value proposition has become more relevant as AI-powered apps increasingly need to talk to many external services at once. The platform manages OAuth flows, API key handling, and token refresh across 800+ APIs, so developers can write integration logic rather than authentication infrastructure. What makes it notable for current workflows: Nango's AI builder generates typed integration functions from natural language descriptions, and the platform works directly with Cursor and Claude Code. Companies like Replit and Ramp use it in production, which gives it more real-world validation than most developer tools at this star count tend to have. It's fully open-source and self-hostable under the Elastic License.
Also Trending
MoneyPrinterTurbo (64,600+ stars) continues its climb as one of the most-watched AI video tools on GitHub — it generates short-form social video end-to-end from a text prompt, handling script, visuals, voice-over, and editing without a video editor in sight. And kortix-ai/suna is drawing attention as an open-source "company OS" where multiple AI agents share a persistent Linux sandbox, institutional memory, and 3,000+ integrations to run autonomous business workflows around the clock.