GitHub trends for September 20, 2026
What was taking off on GitHub on September 20, 2026: peak magnitude of the day, summary and where it was spotted.
-
9.2
browser-use/jev-ultrafast
A browser agent with a dynamic, indexed action space: given one natural-language goal it picks an operation (CLICK, TYPE_TEXT, SELECT, etc.) and an element, using a small LLM to generate text only for typing.
-
9.1
NandhaKishorM/laya
Laya is a non-autoregressive decision engine that answers typed questions (choice, score, noul) over any state (text, email, ticket, JSON) in a single forward pass (~35 ms on GPU) with calibrated probabilities.
-
8.9
cloudflare/security-audit-skill
A coding-agent skill that turns the agent into a security auditor, running a multi-phase code audit with independently verified, machine-readable findings.
-
8.9
robbietilton/Compositor
A free, open-source raster image editor for macOS written in Swift — a Photoshop alternative focused on compositing and post-processing.
-
8.3
eternity4719/HowToLiveBetter
An evidence-based life guide: 498 recommendations on longevity, first aid, personal finance, law, employment, family and emigration. Each lists costs, benefits, evidence level and primary sources.
-
8.3
mizorewww/laya-mlx
Native MLX runtime for Laya typed decision models: returns probabilities for choices, scores or truth values without text generation, running locally on Apple Silicon.
-
8.3
trycua/cua
Open-source toolkit for computer-use agents: cloud and local desktops, an app automation driver, and benchmarks for training and evaluation. For those giving AI agents control of a computer.
-
7.8
bespokelabsai/nimble
Nimble is a model and training recipe for fast typed decisions over text: given a flat schema of enum and boolean fields, it picks an answer and returns probabilities for each option. It targets routing, condition…
-
7.6
Open-Dev-Society/OpenStock
An open-source alternative to paid market platforms: real-time price tracking, personalized alerts, and company insights. Built with Next.js, MongoDB, and Finnhub, free for everyone.
-
7.4
TianyuCodings/NanoJev
A nano replica of Jev built on Qwen3-0.6B that outputs probability distributions over candidates in a single forward pass with no token decoding, plus a training pipeline and game demos (maze, Snake).
-
7.4
typesafe-ai/skills
Agent skills for building with TypeSafe's System One API: typed decisions and probabilities for LLM agents. Installs as a Claude Code plugin or via skills.sh.
-
7.3
jackwener/wx-cli-again
A Rust CLI for querying local WeChat data: sessions, chat history, search, contacts, group members, favorites, stats and export. It decrypts local SQLCipher databases and returns JSON output designed for AI agents.
-
7.3
rmalde/minecraft-agent
An AI agent that plays Minecraft Java 1.16.5: the Astra GPT planner sets objectives, JEV picks actions, and Mineflayer executes them via the normal player protocol. Includes run recording and verification.
-
7.2
higgsfield-ai/higgsfield
Fault-tolerant GPU orchestration and ML framework for distributed training of models with billions to trillions of parameters, including LLMs. Manages node access, experiment queueing, and GitHub Actions integration.
-
7.1
githubnext/localjev
A local TypeScript/Bun HTTP service implementing a Jev-compatible POST /v1/systemone API on top of an OpenAI-compatible Chat Completions endpoint (DiffusionGemma via oMLX). It acts as a bridge for typed decisions…
-
7.1
jaredpalmer/kev
kev is a LoRA adapter with a small readout head on top of Qwen2.5-0.5B that answers many typed questions about a document in a single forward pass, returning calibrated probabilities instead of text.
-
7.0
docling-project/docling
Docling is a document parsing library for many formats (PDF, DOCX, PPTX, XLSX, HTML, EPUB, audio, video) with advanced PDF understanding: page layout, tables, formulas, OCR. It prepares structured data for generative…
-
7.0
Haleclipse/CometixCode
An unofficial Rust reimplementation of Anthropic's Claude Code terminal UI on top of iocraft, porting the original file by file. Work in progress: large parts of the interactive loop, tools, permissions, MCP and slash…
-
7.0
latent-spaces/brag
An agent skill (Claude Code, Codex, opencode, etc.) that turns a project you just built into a short launch video with music, motion, and share copy — in one command. Rendering is powered by Hyperframes.
-
7.0
tamaratran/fast-jev-compaction
Claude Code plugin that replaces context compaction summaries with Jev decisions: every tool call and result is scored in one fast request, stale ones are dropped or truncated, everything kept stays verbatim. Also…
-
6.9
TheoLeeCJ/SemIf-OpenJev
An open reproduction of the Jev-style semantic decision interface: reads typed option probabilities directly from a 4B model's logits without generating text. Runs on a single RTX 3090.
-
6.9
vercel-labs/json-render
Generative UI framework: an LLM turns prompts into JSON interfaces constrained to a catalog of components you define, keeping output predictable and safe. Supports React, Vue, Svelte, Solid, React Native, Next.js, 3D,…
-
6.8
FareedKhan-dev/train-llm-from-scratch
Educational project: a from-scratch PyTorch Transformer plus a full LLM training pipeline — from raw text through SFT, reward modeling, PPO/DPO/GRPO to chat, without transformers, trl or peft.
-
6.6
coder/coder
Self-hosted platform for cloud development environments and AI coding agents: workspaces are defined in Terraform, connected via a Wireguard tunnel, and shut down automatically when idle.
-
6.6
jarrodwatts/jev-trader
Trading bot for the Kuru DEX on Monad: every block (~300 ms) an AI model decides buy or sell from the MON-USDC order book and posts a post-only limit order one tick inside the touch, earning the spread. Includes a…
-
6.6
Sac-Y/Jev-cu
Jev-cu connects Computer Use with the Jev model (TypeSafe System One): Jev picks the element, action, completion level and risk from textual UI candidates, while Codex Computer Use reads the interface and executes…
-
6.5
arvindear/wp2shell-PoC
Proof-of-concept for the wp2shell vulnerability chain in WordPress Core (CVE-2026-63030 and CVE-2026-60137) that gives an unauthenticated attacker RCE. A tool for testing and exploitation in controlled environments and…
-
6.5
dofastted/vm2api
A proxy service that turns a Setup Token into a Console-compatible Anthropic API: requests run through VMs with a native Claude Code subagent, without prompt injection. For those wanting to use Claude via the standard…
-
6.4
hypit-ai/hypit
Hypit gives AI agents a language and system to create video: the agent clones a viral clip as an editable workflow — footage, captions, B-roll and effects anchored to words instead of seconds.
-
6.4
incoai/splash
A local LLM inference engine for Apple silicon, specialized per model: serves OpenAI- and Anthropic-compatible APIs to coding agents on a single Mac.
-
6.4
mizorewww/laya-coreml
Local port of the Laya model to Apple Core ML and Neural Engine: returns typed decisions (choice, score, yes/no) without token generation, with speed and energy benchmarks.
-
6.4
yibie/awesome-jev
A curated list of public projects, integrations, and discussions built on Jev — TypeSafe AI's System One model for typed decisions (choice, score, boolean with confidence).
-
6.3
awlevin/typesafe-computer-use
Drives macOS toward a goal typed in plain English: OCR the screen, pick the next action via a TypeSafe classifier, click. Costs about $0.0002 per step and never sends screenshots to large models.
-
6.3
featherless-ai/simple-jev
A Python server that turns open Hugging Face LLMs into a classifier: it reads next-token logits and returns JSON with choices, rubric scores, or truth judgments without training a separate classifier head.
-
6.3
HyNetworks/OpenGFW
An open-source DIY Great Firewall for Linux: it reassembles IP/TCP flows, analyzes protocols (HTTP, TLS, QUIC, DNS, SSH, VPNs) and filters connections via an expr-based rule engine. Aimed at home routers and servers…
-
6.3
ruanyf/weekly
A weekly tech digest by Ruan Yifeng: articles, tools and resources, published every Friday. Also includes a section with programmer job postings.
-
6.2
anthropics/financial-services
Reference agents, skills, and MCP connectors for financial-services workflows: investment banking, equity research, private equity, and wealth management. Installs as a Claude Cowork plugin or deploys via the Claude…
-
6.2
BuilderIO/agent-native
An open-source TypeScript framework for building agentic apps where each capability is defined once as an action, used by the agent as a tool and called by the UI from code. Aimed at developers who need agents paired…
-
6.2
QwenLM/Qwen-Image-2.1
Qwen's open-source text-to-image generation and image editing model, including transparent RGBA output and up to 10 reference images. 7B generation component, 2K support, integrations with Diffusers, ComfyUI, vLLM,…
-
6.1
gylive/ccodex-sleep-state
A local Go tool that tries to improve Codex behavior: it collects and injects X-Codex-Turn-State and manages proxies, subscriptions and Astra/Sol/Terra models via a web panel. It does not guarantee better model quality…
-
6.1
joeseesun/qiaomu-download
Agent skill that downloads videos and extracts audio/subtitles from YouTube, Bilibili, X, TikTok, Vimeo, Instagram, WeChat Channels and other yt-dlp-supported links via a single phrase 'download this: URL'.
-
6.1
kerpopule/hermes-jev-skills
A set of eight SKILL.md files and a CLI that wires the Jev decision model into Hermes, Claude Code and Codex agents for model routing, memory filtering, skill selection and GUI actions.
-
6.0
cloudflare/quiche
A Rust implementation of the QUIC transport protocol and HTTP/3 with a low-level API: the application handles sockets, event loop and timers itself. Used in Cloudflare's edge network, Android's DNS resolver and curl…
-
6.0
sutro-sh/jev-align
An experimental CLI from Sutro for building AI Functions with TypeSafe's Jev: it finds ambiguous examples in your dataset, asks you to label them, and uses GEPA to improve the function.
-
6.0
Tencent/BrowserSkill
BrowserSkill connects AI agents (Cursor, Claude Code, Codex, etc.) to your already logged-in browser via the bsk CLI and an extension, letting the agent work in a separate window without interrupting you.
-
6.0
unicodef1wn/grokbot-field-notes
Rules, playbooks, bot roles and a failure log from the xAI Grok Bot team's 72-hour live build: drop AGENTS.md into your repo root and your agent reads it.
-
5.9
mihail911/modern-software-dev-assignments
Assignments for Stanford's CS146S course "The Modern Software Developer", covering modern software development practices with Python 3.12, Conda and Poetry.
-
5.8
anthropics/knowledge-work-plugins
A collection of open-source plugins for Claude Cowork and Claude Code that turn the assistant into a specialist for a specific role: sales, support, finance, legal, marketing, data and more. Each plugin bundles skills,…
-
5.7
ethanplusai/codex-on-crack
A Codex skill that splits coding work between two models: Astra plans, reviews and makes decisions, while DeepSeek V4.1 Flash handles bulk implementation, testing and debugging. Aims to cut the cost of expensive model…
-
5.7
ghuntley/underclass
An OpenAI-compatible local proxy that pools multiple ChatGPT/Codex and GitHub Copilot subscriptions behind one endpoint, with sticky sessions, quota cooldowns and fail-fast behavior.
-
5.7
huangbai-AI/post-production-skill
A Codex skill that rewrites ideas or reference material into cinematic video prompts for Seedance 2.5, covering VFX, 3D UI, transitions and camera movement.
-
5.6
ejfkdev/ddc
A Rust DEX-to-Java decompiler for Android with progressive analysis: query metadata first and decompile classes on demand instead of a full run. Its output is javac-verified across seven real-world APKs.
-
5.6
Heman10x-NGU/openJev-verdict-2.0
An open non-autoregressive decision model built on ModernBERT (149.6M) that emits typed decisions (Choice, Score) with calibrated probabilities in ~20-25 ms. On the typed-decisions benchmark it reaches 77.10% accuracy…
-
5.6
rustfs/rustfs
RustFS is a distributed object storage system written in Rust with an S3-compatible API, positioned as a faster Apache 2.0 alternative to MinIO. It is optimized for data lake, AI, and big data workloads.
-
5.6
shhivv/third-hand
A small macOS menu bar assistant: focus an app, press Control–Space, and tell it what to do. It reads accessible controls, types, clicks, and checks the result using Accessibility, Apple Vision, and the Jev model.
-
5.6
yynxxxxx/Codex-X
Cross-platform GUI tool for OpenAI Codex (desktop/CLI): manage providers and APIs, prompt templates, sessions, Skills/MCP and TOML config without hand-editing files.
-
5.5
anthropics/claude-code
An agentic coding tool that lives in the terminal: it understands your codebase, executes routine tasks, explains complex code, and handles git workflows via natural language commands.
-
5.5
guohuiyuan/go-music-dl
A Go-based music search and download tool for 10+ platforms (NetEase, QQ, Kugou, Bilibili, etc.) with Web, TUI and desktop modes, supporting playlists, albums and lossless FLAC.
-
5.5
Mak5er/AirCard-iOS
iOS app that writes custom artwork into iOS 27 system caches over a local loopback tunnel, letting you restyle Apple Wallet cards and the lock screen passcode keypad directly on the iPhone.
-
5.5
MelisaPeteRs2006/Silent-Crypto-Miner
A cryptocurrency mining builder and runtime with CPU/GPU support, multiple algorithms and pools, plus a stealth mode (process disguise, auto-start, pause on activity).
-
5.5
v-modal/awesome-jev-tools
A curated list of public projects and practices built on Jev, TypeSafe AI's System One model for typed decisions that takes unstructured state and returns a typed answer with a confidence rating. It gathers production…
-
5.4
addyosmani/agent-skills
A collection of 25 engineering skills and 9 slash commands for AI coding agents, covering the development lifecycle from spec to ship. Helps agents follow proven practices: TDD, code review, task planning.
-
5.4
browserbase/stagehand
SDK in TypeScript, Python and Go for LLM-driven browser control: observe/act/extract find selectors, perform actions and return schema-validated data. Built for agents that operate websites like humans.
-
5.4
ERRRC/kaogongzhentizhengliu
An Obsidian vault of 22,375 notes reorganizing 2016–2026 Chinese civil service exam questions by topic, with deep annotations, links and local images.
-
5.4
mindbox77/zxdesk
A graphical windowed OS for the 48K ZX Spectrum written in Z80 assembly: overlapping windows, menus, a heap, an event queue, a file manager and settings — all fitting in 48K and running on real hardware.
-
5.4
razorback16/openjev
An open-source "System One" decision server on DiffusionGemma: send a state and typed questions (yes/no, choice, score) and get probabilities plus confidence in tens of milliseconds, with no text generated.…
-
5.4
TencentCloud/Octop
Self-hosted AI assistant in Python with multi-user and multi-agent support: a single process serves a web dashboard, CLI, IM integrations and cron, with all data kept locally.
-
5.3
ahmedkhaleel2004/gitdiagram
A web service that turns any public or private GitHub repository into an interactive architecture diagram: it analyzes the file tree, README and code excerpts with an LLM to build a system graph with links to sources.
-
5.3
cobanov/awesome-jev
A curated list of projects built with Jev, TypeSafe AI's System One model for fast typed probabilistic decisions (classification, routing, scoring, guardrails).
-
5.3
krillinai/OpenCreator
Open-source AI workspace for creators powered by Codex CLI: visual tools and agents for video, images, voice, avatars, translation, and editing in one place.
-
5.3
kyotofin/tax-doc-classifier
A TypeScript tax document page classifier that identifies one of 261 IRS forms and the page kind from page text using the Jev decision model instead of an LLM. No training or hosting required — just a JSON file of form…
-
5.3
lakeday-org/perch
CLI linter that finds semantic defects in code (wrong call order, inverted conditions, off-by-one) and ranks them by probability. Supports custom rules in perch.yaml and integration with agents (Claude Code, Codex,…
-
5.3
Lumid-Off/AirCard-Windows
Native Windows app in Rust for customizing Apple Wallet cards and lockscreen passcode themes on iOS 18+ without jailbreak via the AirTraffic sync exploit.
-
5.3
minorun365/minorun-marp-skill
A skill set for AI agents (Claude Code) and humans for building conference slides with Marp: rules for storytelling, figures, and a dark theme, plus scripts to inspect PDF and SVG output.
-
5.2
cactus-compute/needle
A 2-bit, 8-29 MB foundation model for phones, wearables, robots and microcontrollers that handles tool calls, structured extraction and text embeddings on-device.
-
5.2
ThreeDaPrint/niimbot
Firmware patch for the Niimbot B1 printer (version 5.22) that enables real density control and full darkness on any label, including third-party media.
-
5.2
wfzyx/von
An open-source non-autoregressive System One decision model for classification, intent routing, and guardrail validation without text generation, with calibrated probabilities and sub-25ms latency.
-
5.2
wuyoscar/jev-skill
A collection of skills and scenarios for Jev, a service that classifies and scores candidate decisions while your agent supplies context and does the work. Installs into Codex, Claude Code or OpenCode as a set of…
-
5.2
zhouxiaoka/autoclip
AI-powered video clipping tool that downloads videos from YouTube and Bilibili, analyzes content with Qwen LLM, and automatically extracts highlight clips and generates compilations.
-
5.1
asciimoo/hister
A private Go search engine that indexes the full contents of visited pages and local files. Search is available via web UI, terminal, or an AI assistant through MCP.
-
5.1
brainstormity/Jev-X-Sentiment-Analysis
A crypto market intelligence terminal that combines live price data, funding rates, and Twitter sentiment to produce a Buy/Sell/Hold decision card with entry, stop-loss, and target levels. It does not execute trades…
-
5.1
hirotomasato/yowes
Python MCP server that generates PNG teacher documents (ID cards, licenses, letters, payslips) for 13 countries. Designed to be called by AI agents over stdio.
-
5.1
liushunqi8-hash/editaplot2026
EditaPlot is a Codex skill for Windows that turns experimental data tables into editable Origin (OPJU) figures with PNG, PDF, and TIF exports, confirming each column's role with the user.
-
5.1
Oldcircle/geo-sleuth
An agent skill that geolocates photos using OpenStreetMap geometry, elevation data, satellite imagery and street view, and shows its work. Works with Claude Code, Codex, Cursor, Gemini CLI, OpenCode and GitHub Copilot.
-
5.1
op7418/guizang-product-video-skill
A skill for AI coding agents (Claude Code, Codex) that turns a product's codebase into a release promo video: reuses real components, writes the script, generates music and SFX, and renders an MP4.
-
5.0
alibaba/open-code-review
AI code review tool from Alibaba: reads git diffs and, via an LLM agent with file access and repo search, produces line-level review comments. Ships with built-in rules (NPE, thread-safety, XSS, SQL injection) and a…
-
5.0
logan-markewich/jeff
A self-hosted drop-in replacement for TypeSafe's jev System One API, powered by the GLiFormer model (400M) and compatible with the official typesafe-sdk. It classifies text locally into choices, ordered scores, and…
-
5.0
logicrw/awesome-jev-projects
A directory of open-source projects built around TypeSafe Jev, a fast typed-decision (System 1) architecture for autonomous agents. It catalogs 260+ projects across 17 categories with commit-pinned source evidence.
-
5.0
superagents-lab/jev-search
A web search app where TypeSafe's Jev interprets plain-language questions to pick sources, time ranges and queries, then ranks results from Search1API with visible relevance scores. Returns links and snippets instead…
-
5.0
tamaratran/jev-pruner
A Claude Code plugin that uses TypeSafe's Jev to trim noisy Bash output after the command runs but before the result reaches the main LLM. It reduces data carried into later turns without generating a summary.
-
5.0
Tencent/WeVisDoc
WeVisDoc is an end-to-end document parser built on Qwen3-VL that converts page images into structured Markdown with LaTeX formulas and HTML tables. 2B and 4B versions are available, served via vLLM.
-
5.0
unclebob/uml-viewer
Live Quil UML class-diagram viewer that lays out topology from an EDN IR and overlays metrics (CC, coverage, CRAP) from .metrics/ snapshots.
-
5.0
ZuodaoTech/everyone-can-use-english
An English-learning project powered by AI: the Enjoy web app, a browser extension for YouTube and Netflix, and materials from the book "Everyone Can Use English".
-
4.9
alicasuu/llmwiki-operational
A local FastAPI + SQLite system that turns technical manuals and SOPs into persistent wiki pages with citations, revision history, and claim-to-source tracing.
-
4.9
Continuum-AI-Corp/OrcaBonsai-27B-Uncensored
A tool for runtime refusal ablation in the compressed Ternary Bonsai 2 27B LLM: it projects the residual stream to remove the refusal direction without modifying or re-quantizing weights. Runs on Apple Silicon via MLX.
-
4.9
deepseek-ai/deepseek-harness
Open-source agent harness by DeepSeek AI written in TypeScript, built on an everything-is-a-plugin architecture powered by Cordis. It runs as a web UI via npx and serves as a base for composing AI agents from plugins.
-
4.9
isas1/skills
A set of two Claude skills: eli5-succinct gives a short plain-language explanation right in chat, while summary builds a self-contained HTML page visually summarising a conversation, branch or diff.
-
4.9
JamesKInner/Covalent-MAS
A modular Python architecture for iterative covalent drug design: molecule generation, filtering, AutoDock Vina docking, multi-objective optimization, and evidence-driven trajectory learning.
-
4.9
kvmem/kvmem-llama.cpp
A llama.cpp port with tiered KV memory: a bounded GPU KV working set, host-RAM storage and query-based retrieval of relevant blocks. Runs Qwen3.8-27B with a 256K logical workspace on 16 GiB VRAM for long-running agent…
-
4.9
LSPosed/ZygiskNext
Standalone implementation of Zygisk that provides Zygisk API support for KernelSU and replaces Magisk's built-in Zygisk. Aimed at rooted Android users on KernelSU or Magisk who need Zygisk module support.