AI Open Source Radar — July 24, 2026

Coverage: GitHub repositories surging in the last 24 hours, ranked primarily by daily momentum (Trendshift daily) and cross-checked with 7-day growth (findarepo). Freshness verified: Trendshift’s top live-mention snowflake ID decodes to 2026-07-23 13:59 UTC (~7 hours before this run), confirming the feed is not cache-stuck. Yesterday’s picks (koala73/worldmonitor, block/buzz, bojieli/ai-agent-book) were dropped after their daily ranks fell sharply.

petergyang/no-ai-slop — Trendshift Daily #2 (new breakout, New 2026)

no-ai-slop is a writing-editing skill that catches and strips 20+ patterns of AI “slop,” and runs like a slash command inside Claude Code, Cursor, and Codex. Paste in your text and it detects and removes formulaic patterns such as binary contrasts (“It’s not X. It’s Y.”), throat-clearing openers (“Here’s the thing”), faux-insight setups (“What nobody tells you”), colon reveals (“The best part: it learns.”), and fake-profound endings (“The future isn’t coming. It’s already here.”), while enforcing fundamentals like leading with the point, active voice, and concrete numbers. It can be used to detect slop as well as remove it, and it assumes a philosophy of drafting by hand and using AI only to edit. Unlike a generic grammar checker, it targets “AI-ness” itself. Creator Peter Yang open-sourced it on July 22–23 on X, and it rocketed to #2 on Trendshift’s daily board amid growing “AI slop fatigue.”

Practical use: Drop it in as a final editing pass to scrub the “AI feel” from blogs, newsletters, and marketing copy and protect brand credibility.

Tags: #AIWritingDetection #EditingSkill #ClaudeCodeSkill #ContentOps #ProseStyle
Repo: github.com/petergyang/no-ai-slop

ZhuLinsen/daily_stock_analysis — Trendshift Daily #7, #1 Python (~49,000 ⭐)

An LLM-powered stock-analysis system spanning multiple markets (China A-shares, Hong Kong, and U.S.). It pulls quotes, real-time news, technical indicators, sentiment signals, and fundamentals from free data sources such as AkShare, Baostock, and YFinance, then has an LLM synthesize a daily “decision dashboard” and auto-push it to WeChat Work, Feishu, Telegram, Discord, Slack, or email. It runs on a schedule at zero cost and lets you swap among multiple LLM providers. Its differentiator is being a self-hosted research pipeline an individual can run without a paid data subscription. Already popular with roughly 49k stars and 43k forks, it reignited today as the #1 Python repo on Trendshift’s daily board.

Practical use: Adaptable as a template for automating stock screening and morning research briefings (not investment advice — informational only; trading decisions remain the user’s responsibility).

Tags: #EquityResearch #MultiMarketData #DecisionDashboard #SelfHostedFinance #ScheduledAutomation
Repo: github.com/ZhuLinsen/daily_stock_analysis

agentlas-ai/Agentlas-OS — Trendshift Daily #12 (new breakout, New 2026)

Agentlas-OS is a local-first “agent OS” that keeps specialist agents in a hub and spins up a temporary orchestrator for each task. From a macOS “control room” it runs teams of local AI agents; when a task arrives, a one-off orchestrator summons only the specialist agents it needs, then tears down when done. It integrates with Claude Code, Codex, Gemini, and BYOK (bring-your-own-key) runtimes, so it isn’t locked to any single model. Its core differentiator is a “per-task orchestrator” design that conserves resources and context instead of keeping a large multi-agent swarm always running, positioning it as a no-code alternative to CrewAI, LangGraph, and AutoGen. The “one sentence → an agent team” no-code approach drew attention and pushed it to #12 on the daily board as a new entrant.

Practical use: Automate repetitive work — research, documentation, code review — as local team-based agents without exfiltrating data.

Tags: #AgentOrchestration #LocalFirst #NoCodeAgents #TaskScopedOrchestrator #BYOK
Repo: github.com/agentlas-ai/Agentlas-OS

ruvnet/RuView — Trendshift Daily #13 (WiFi sensing back in focus)

RuView turns ordinary WiFi signals into real-time spatial intelligence, vital-sign (breathing and heart-rate) monitoring, and presence detection — without using a single camera pixel. A roughly $9 ESP32 board reads the radio reflections off people in a room, and a small pretrained model infers who is present, how they are breathing, and how their heart rate is trending. It works by capturing the subtle ways people reflect, refract, and absorb WiFi waves as they move, building a real-time map of spatial activity; the 4-bit quantized model is just 8 KB and runs in microseconds on a Raspberry Pi. Its differentiators are detection through walls and in the dark with no wearables, plus native integration with the major smart-home ecosystems — Home Assistant (MQTT), Apple Home (HAP), and Google/Alexa (Matter). “Video-free, privacy-preserving sensing” brought it back into focus across smart-home and health communities, lifting it to #13 on the daily board.

Practical use: Low-cost contactless-sensing PoCs such as elderly fall/breathing monitoring and office occupancy/energy optimization.

Tags: #WiFiSensing #ContactlessVitals #EdgeInference #PresenceDetection #SmartHomeIntegration
Repo: github.com/ruvnet/RuView