09.26 Today's Picks

357 new projects found today, 96 rated.

Created 2026-09-26 – 2026-09-26Rated 109

A · Worth watching

interference-search

BadtheorylabsPython

Lets a model reason in parallel across multiple explicit state branches, merging duplicates, pruning dead ends and advancing together, solving agent search divergence; 93 stars in 0.4 days is a strong signal. It has a path to becoming the standard layer for agent reasoning, with the risk of overlapping Tree-of-Thoughts implementations and mainstream agent frameworks, so it needs to ship benchmark data soon.

Reason over explicit states, many branches at once: merge duplicates, cancel dead ends, advance together.

#AI Agents #LLMs

★93

phantomat

kaoltiC++

Infinite-canvas window management for Hyprland: windows become zoomable, pannable layers rather than tiles in a grid, a new way to organize research, writing, and multi-screen comparison. 158 stars in 1.7 days shows real demand; the limits are that it only serves Hyprland users and stability is unverified.

Phantomat: a zoomable, infinite-canvas window manager for Hyprland

#Linux #UI Design #3D #hyprland #wayland

★158

godot-lunastream

xrarlenalZig

Lets Godot 4's VideoStreamPlayer play RTSP and http streams directly, with FFmpeg demuxing plus hardware decode, GPU zero-copy, and no external runtime, filling a long-missing engine gap. Godot developers doing live casting or surveillance feeds can use it immediately; the risk is unverified GDExtension prebuilt coverage and stream protocol boundaries.

Godot 4 GDExtension: play rtsp:// / http(s):// / local files through VideoStreamPlayer — FFmpeg demux, hardware decode, GPU zero-copy, no external runtime.

#Game Dev #Video #Streaming #godot #ffmpeg

★21

servery.nvim

wurliLua

A small Neovim plugin that uses an existing fuzzy finder to jump between multiple server sessions, doing only that with an honest description. Immediately installable convenience for people running several nvim instances, but several session-switching plugins already exist, the scope is narrow, and it's hardly a new approach.

Jump between Neovim sessions using your favourite fuzzy finder

#Plugins #Editors #Productivity #neovim

★24

open-beni

Mondo-RoboticsTypeScript

It brings robot racing to the browser: MuJoCo compiled to WASM, ONNX policies running online, plus training source, more complete than most robot demos. Suited to those trying reinforcement learning and physics simulation cheaply, though it's a toy and far from a real robot stack.

OpenBeni browser robot racing with MuJoCo WebAssembly, ONNX policies and training source

#Robotics #WebAssembly #ML Research #mujoco

★19

cdj-nxs2-qemu

Stijn-JacobsC

It runs Pioneer CDJ-2000NXS2 deck firmware under QEMU, a rarely touched corner of hardware emulation, interesting to those reverse-engineering DJ gear or wanting to test without the physical unit. The limit is that it's purely technical experimentation, and it's unclear whether a full playback chain can complete.

CDJ-2000NXS2 QEMU Emulator

#Embedded #Reverse Engineering #Hardware #qemu

★18

cida

XuanwoSwift

A small always-on macOS tool that calls a chosen LLM to translate and polish text in any input field; scoped, accurately described, install-and-go, from an author with a track record in open source. Suited to people who write English often, though similar Raycast plugins already cover most of the ground.

Cida: translate and polish text anywhere on your Mac using the LLM of your choice. Translate and polish text anywhere on your Mac.

#macOS #Productivity #LLMs #Editors

★26

That's today. The other tiers and the boards follow.

BHas a spark17

B motion-video-skill An agent skill that one-click produces 1080p videos with AI voiceover, subtitles, and generated music, directly matching content creators' needs; 47 stars in 0.7 days and an MIT license aid diffusion. It could monetize through skill subscriptions or API billing, but the risk is replacement by one-click video products like Runway and Pika and many free similar skills. #Agent Skills #Video #Animation

Agent skill that produces beat-synced 1080p motion-graphic videos in HyperFrames (HTML + GSAP) with an AI voice-over, karaoke captions, SFX and generated music. MIT.

bestagentkitsJavaScript★47
B json-guard Limits parse depth, key count and length to reject JSON bombs before allocation, directly closing the DoS attack surface; 51 stars in 1.6 days shows security and backend teams can use it immediately. It could become the default secure parsing library in the Node ecosystem and monetize via compliance audits, with the biggest risk being Bun/Node native protections or zod competitors absorbing the capability. #App Security #Performance

JSON.parse with hard limits on nesting depth, key count, string length and input size. Refuses parser bombs before they are allocated.

agentscope-ai-javaTypeScript★51
B omnibin Puts every binary in nixpkgs history directly on PATH, solving the core friction of wanting to use Nix immediately; 76 stars in 0.9 days is strong momentum. It has potential to become a standard tool for Nix users with monetization via hosting or support, but the risk is that official nix shell/flake experience improvements gradually erode its necessity. #CLI #Build

Every binary nixpkgs ever shipped, on your PATH.

fzakariaPython★76
B unreal-home-wizard Uses Codex to turn room and home photos into a walkable, reviewable UE project, automating the path from photography to 3D scenes; 49 stars in 1.7 days is decent momentum. It could charge per use for real estate showcases and game prototyping studios, but the risk is quality depending on LLMs and UE version compatibility, with image-to-3D services like Meshy as strong substitutes. #3D #Game Dev #AI

Turn room and home photographs into reviewable, walkable Unreal Engine projects with Codex.

amirmushichgePowerShell★49
B pg-outbox A driver-free, dependency-free Postgres transactional outbox that plugs in with any query function, cutting distributed consistency boilerplate; 20 stars in 1.6 days is normal growth. There's a path to becoming a Node standard dependency with monetization via a hosted service, but the risk is that Debezium/CDC and logical replication replace the outbox pattern. #Databases #APIs

Transactional outbox for Postgres with no driver dependency — pass in any query function.

agentscope-ai-javaTypeScript★20
B agentic-cuda-optimizer A langgraph workflow drives an agent to run the CUDA harness, read benchmarks and nsight profiles, and automatically modify kernels, closing the performance tuning loop; 27 GitHub stars plus 36 HN points gives dual-source momentum. It could be a performance optimization SaaS or embedded in CI, but the risk is poor reproducibility of tuning results and competition from NVIDIA's auto-tuning tools. #AI Agents #GPU #Performance

A langgraph based workflow with a C++ CUDA harness to optimize CUDA kernels

bertayePython★27
B umbod Lets teams connect internal APIs and MCP on their own infrastructure and manage agent permissions by group, hitting enterprise agent governance and data egress pain points; 20 stars in 1.6 days but lacking customer evidence. The enterprise subscription path is clear with high deal size, and the risk is that model gateways and identity vendors are also extending into the permission layer. #MCP #AI Agents

Agent access on your own infrastructure. Connect internal APIs and MCP servers, manage permissions by group, and let teams choose their agents.

computerlovetechPython★20
B ai-system-design Step-by-step explanations of LLM, RAG, and Agent-based system design, filling a gap in interview and architecture learning material; 154 stars in 0.8 days is extremely strong momentum, proving real content demand. It can monetize through courses, consulting, and a newsletter, but the risk is that content repos have a low ceiling and are easily replaced by top education institutions' content. #Learning #LLMs

AI System Design - Learn how to design AI systems built on LLMs, RAG, and AI Agents step by step.

amitshekhariitbhuMarkdown★154
B collabosm Gives a reproducible recipe and measured throughput for running a 125B MoE on a single Colab A100, directly solving the cost of personal large-model experiments; 70 stars in 1 day is strong momentum. It could become a reference implementation for inference deployment and funnel to hosted services, with the risk that weights and runtime versions expire fast and vLLM official docs easily supersede it. #LLMs #Performance #GPU

Run Qwen3.8-Flash-Next (125B-A6B MoE) on ONE Colab A100-80GB High-RAM: pinned runtime, weights from HF, OpenAI-compatible endpoint, measured 2.8k-3.9k t/s prefill / 90-97 t/s decode, plus a verified p…

architectdsPython★70
B motion-graphics Packages motion graphics as a skill Claude Code and Codex can call directly, riding the agent-skill ecosystem boom; 29 stars in 0.5 days is decent momentum. It can monetize through skill marketplaces or paid skills, but the risk is that the official skill ecosystem and many free similar lists quickly dilute its value, and it lacks exclusive capabilities. #Agent Skills #Animation

Motion-graphics skills for Claude Code and Codex.

Barty-BartHTML★29
B flux Connects an Omarchy host's files, clipboard, notifications, media, and camera to the phone, solving the desktop-phone split; 46 stars in 0.2 days is a very good initial pace. It could monetize via open-source plus hosted dual tracks, but the risk is that KDE Connect, LocalSend, and Snapdrop already cover most functions, leaving differentiation and willingness to pay unproven. #Productivity #Automation

Connect an Omarchy computer to your phone: files, clipboard, notifications, media, camera, and more

bjarneoKotlin★46
B ft-motion Treats motion graphics as a pure function of time, rendering headless Chrome and compositing with ffmpeg, plus a brief-to-video prompt for coding agents, filling a gap in agent-produced video; 20 stars in 0.4 days. It could charge via skills or API, but risks competition from After Effects scripts and similar agent video skills. #Agent Skills #Video

Motion graphics as a pure function of time — Canvas 2D scenes, headless-Chrome rendering with motion blur, ffmpeg, synthesized sound, and a brief-to-video prompt for coding agents.

imserhatdemirJavaScript★20
B ComfyUI-Fisher-Pose A pose and camera editor in ComfyUI for Qwen Image portrait generation, with a built-in skeleton library and free multi-camera views, addressing exactly the pain of uncontrollable composition in AI images. Directly usable for illustration workflows, though tightly bound to a specific model and plugin ecosystem. #Image Generation #Plugins #Images #comfyui #openpose

ComfyUI pose and camera editor for Qwen Image 2.1: VNCCS realistic human figures + one-click OpenPose posing (built-in 209 skeletons) + free-view multi-person camera | Pose & camera studio with VNCCS …

Work-FisherJavaScript★32
B VeriTile A Lean 4 framework for formally reasoning about Triton-style kernels, targeting the nearly untouched corner of GPU operator correctness verification, meaningful to researchers in compilers or formal methods. The limits are a very high barrier and unclear use cases; at this stage it's more paper companion than usable tool. #ML Research #GPU #Testing #triton

Lean 4 framework for reasoning about Triton-style kernels.

Lizn-znLean★17
B keelflow A community continuation of Flowise: it keeps only the Apache-2.0 code and focuses on security fixes, with existing flows and data migrating directly. It does save effort for teams with compliance needs who want visual agent orchestration, but it's still a low-code LLM platform with limited innovation, and maintenance effort needs checking. #AI Agents #LLMs #Self-Hosted #flowise

Build AI agents and LLM workflows visually. Security-maintained continuation of Flowise: security fixes, Apache-2.0 only, drop-in for your existing flows and data.

PerruerTypeScript★30
B SFSymbolReplacer A small tool that globally replaces macOS system SF Symbols with custom PNGs, doing one thing with a direct name. Suited to theme customization and UI experiments; the limits are the risk of touching system resources and a very narrow audience. #macOS #UI Design #Plugins

Replace system SF Symbols with your own PNGs, system-wide.

mszhangopopopObjective-C★24
B DLSSNR-AMD An attempt to run DLSS neural rendering models on AMD RDNA4 via Vulkan, a hardcore experiment moving Nvidia-exclusive tech to AMD cards, appealing to graphics reverse-engineering fans. The limits are unclear validation and being far from usable, with only a one-line description. #GPU #Hardware #3D #dlss #vulkan

DLSS 5 neural rendering (NR) model on AMD RX 9000 (RDNA4) through Vulkan

mochizuki0323C++★16

COrdinary26

C parquet-schema-check Reads the Parquet footer with zero dependencies to catch schema drift, avoiding crashes only at the reader end, a real pre-check for data engineering; 20 stars in 0.5 days is okay but there's no user feedback. It could be folded into a data quality platform, but the risk is that mature ecosystems like Great Expectations and deequ already cover similar validation. #Databases #Data Analysis

Find incompatible schema drift across Parquet files before a reader hits it. Reads footers with zero dependencies.

agentscope-ai-javaPython★20
C flick-computer-use Provides local browser and macOS computer operation for MCP agents with result validation, hitting the computer-use reliability pain point; 19 stars in 1.8 days is slow. It could be paid agent infrastructure, but the space is crowded, with official Anthropic/OpenAI tools and various open-source harnesses as strong competitors. #AI Agents #MCP

Fast browser and macOS computer use for MCP agents. Local execution, TypeSafe Jev decisions, verified outcomes.

bgivenbTypeScript★19
C OpenGhost A general Windows desktop agent covering system-level operation automation, addressing a real problem; 19 stars in 1.9 days with no HN validation means weak momentum. The commercial path is a computer-use subscription, but the space is already crowded by OpenAI Operator, Anthropic Computer Use, and Open Interpreter, leaving a small independent window. #AI Agents #Productivity #Automation

Open desktop agent for Windows

ANDRETRIPOLJavaScript★19
C cf-clearance Obtains cf_clearance cookies in one shot without a browser to bypass Cloudflare challenges, addressing a real need in scraping automation; 38 stars in 0.3 days is clear momentum. It could charge per use, but it's an adversarial business facing continuous Cloudflare blocking and legal/ToS risk, making it hard to become an accepted standard dependency. #App Security #Reverse Engineering

Get a valid cf_clearance cookie from the Cloudflare Challenge automatically - Python library, CLI and examples, one REST call, no browser required.

Demet5292Python★38
C sqlite-fts5-check Detects inconsistency between FTS5 indexes and content tables, filling a blind spot in integrity-check; the problem is real but extremely narrow, at 20 stars in 0.5 days. It can only supplement the SQLite toolchain, monetizing through maintenance funds or folding into DB tool suites, but the audience is too small and without official adoption it won't become a standard dependency. #Databases #Testing

Detect drift between a SQLite FTS5 index and its content table, which integrity-check does not catch.

agentscope-ai-javaPython★20
C captcha-api Wraps Cloudflare Challenge and Turnstile captcha solving into a paid API with concurrency, batching and cost stats, hitting 130 stars in two days, showing real paid demand from scrapers and ticket-grabbing circles. It would charge per solve, but bypassing protections is a gray area, Cloudflare upgrades can break it anytime, and it faces 2Captcha, CapSolver and legal bans. #Crack #APIs

Captcha-solving API client - Cloudflare Challenge and Turnstile in one interface, with cost tracking and concurrent batch solves.

jamespolyakov9829Python★130
C importmap-lint Validates import maps per the HTML spec algorithm and flags CDN entries without locked versions, targeting supply chain drift; 20 stars in 0.5 days is a decent start. But it's a very niche CI checker easily absorbed by browser warnings or bundlers, with limited paid and platform potential and unlikely to become a team default dependency. #Code Quality #Web Frontend

Validate import maps against the HTML spec resolution algorithm, and flag unversioned CDN entries.

agentscope-ai-javaTypeScript★20
C sakuragaoka-station A walkable, animated cherry-blossom station town generated procedurally in pure three.js; the one-line description is honest and the scope clear, making it a nice 3D scene demo. It's a reference for procedural modeling or cel-shading work, but that's all, and shader toys are the alternatives. #3D #Web Frontend #Animation #three.js

A walkable first-person anime cel-shaded Japanese sakura station town, fully procedural in plain three.js

Kenton-GMIJavaScript★17
C Qpro-Enhanced-FT-Wireless Implements wireless tongue tracking and eye aggregation on a rooted Quest Pro, a hardcore hardware-hacking input enhancement; 41 stars in 1.2 days shows real demand in the VR community. It could monetize toward accessibility input or peripherals, but is limited to a single device model, root risk, and Meta eventually shipping official features, so the ceiling is low. #Hardware #Accessibility

Enables tongue tracking and eye convergence on a rooted Quest Pro over wifi

FwooffyPython★41
C mysticgsi A script tool that builds a Generic System Image from any official firmware, convenient for flashing enthusiasts, but GSI packaging scripts are already plentiful, and 24 stars in 0.2 days looks suspicious. The description is too short, with no supported scope or risks stated. #Android #Build #CLI

A tool to build a Generic System Image from (almost) any stock firmware

MysticGSIShell★24
C cabbage-trading-machine A CCXT-based spot trading bot with RSI plus EMA signals, backtesting, and sim/live switching; usable but the strategy is textbook-level. Similar projects are everywhere, and 170 stars in 1.4 days looks more like riding the quant hype than technical recognition. #Fintech #Crypto #Automation

Spot trading bot - RSI + EMA signals, backtest, paper and live trading via CCXT. Built on Investing Algorithm Framework.

sopersonePython★170
C RaymanOriginsRecomp Uses XenonRecomp to statically recompile the Xbox 360 version to PC/Android, reusing porting technology in a game-preservation engineering effort; 16 stars in 1.1 days is weak. It ships no assets and has unclear licensing limiting commercialization, so it can only rely on community sponsorship or porting services, and it's tied to a single recompiler ecosystem. #Game Porting #Reverse Engineering

Unofficial static recompilation of Rayman Origins (Xbox 360) for PC and Android, using XenonRecomp. No game code or assets included.

BelmanteGuC++★16
C DownMax A macOS downloader plus Chrome/Safari extensions covering YouTube, VK, Instagram and more, broad in features but technically unoriginal, and it specifically supports paid course sites like GetCourse. For the same needs, yt-dlp is more stable and transparent. #Downloaders #macOS #Browser Extensions #Video

DownMax: downloader for macOS and extension for Chrome and Safari — YouTube, VK Video, Instagram, Threads, Vimeo, GetCourse

ohlexlexaSwift★58
C pencil-by-rino Adds a front-end visual editing entry point to AI-generated WordPress themes, a concrete idea with a clear scenario. But 16 stars in 1.6 days and a PHP implementation leave no clear difference from the block editor, and maturity is hard to judge. #Web Frontend #Editors #wordpress

Frontend content editing for AI-built WordPress themes.

rinothecoderPHP★16
C MHI2Q-CarPlay-AltScreen A patch script for Audi MHI2Q/MIB2 head units that projects the CarPlay secondary screen onto the virtual cockpit. Real demand but a very narrow audience, lacking applicable firmware versions, flashing risks, and rollback instructions; it looks like a snippet a vendor casually released, and ordinary owners following it face real risk. #Reverse Engineering #Hardware #Embedded #carplay

Completes CarPlay AltScreen logic for Audi MHI2Q / MIB2 High AUG22, displaying the CarPlay secondary screen on the Virtual Cockpit.

yuedizhiboShell★54
C ntm-next A NeoForge/Fabric port of the HBM nuclear tech mod, a must-have update for Minecraft players of this long-standing industrial mod. To outsiders it's just another large modpack, with little technical novelty, and the version numbers in the description are vague. #Games #Game Dev #minecraft

HBM's Nuclear Tech Mod for Minecraft 26.2 (Fabric and NeoForge)

Warfactory-OfficialJava★30
C truffler A semantic search layer on Rails, adding tags, query understanding, and streaming reranking on top of keyword and vector retrieval. The idea isn't new, and it also flies the Jev banner, highly suspected of sharing origins with that batch of collaboration repos, so credibility is discounted. #RAG #LLMs #Databases #rails

Sniff out the right record: Jev-powered search for Rails with index-time labels, query understanding, and streamed reranking on top of your own keyword and embedding search

kieranklaassenRuby★18
C saas-motion-kit A template pack for mass-producing SaaS promo videos with Claude Code: transition tables, storyboard themes, tone matrices, essentially prompts plus asset collections. Marketing video makers can copy the homework, but it isn't a tool and output easily becomes homogeneous. #Video #Content Writing #LLMs #claude

Promo & motion videos for software products with HyperFrames + Claude Code. No two films should feel the same: tone matrix, variety audit, 24-transition atlas, 100 storyboard themes.

tugrawork-creatorPython★50
C quicksilver A Claude Code skill called "hand judgment to Jev," pitched on saving tokens and speeding things up with benchmark numbers. But the Jev name appears across several new repos at once, looking like a coordinated marketing push, and the benchmarks can't be verified; better to wait than adopt. #AI Agents #LLMs #CLI #jev

Claude Code skill: hand bulk judgment calls to Jev. 86% fewer Claude tokens on a 12-task benchmark, up to 20x faster. One-line npx install.

UditAkhouriiJavaScript★34
C Seep-Tool A white-box audit methodology and hardening framework around the CWE-602 client-side authorization flaw; the topic is specific, but the output is a vague "methodology plus research framework" with no runnable scanner or rule set. 39 stars in 0 days is also doubtful. #App Security #Code Quality #Testing

CWE-602 client-side authentication weakness white-box audit methodology and defense-in-depth hardening research framework

angusdevgoLanguage unknown★39
C lcs-recomp From the name, likely a static recompilation project for Liberty City Stories, 68 stars in 1.1 days, but the repo has no description, build steps, or progress statement. Recompilation projects can least afford a black box; for now it's just a placeholder. #Game Porting #Reverse Engineering #Games

elmasasC++★68
C vllm-jev Integrating the Jev decision model into vLLM's native serving layer is routine adaptation work, and its value depends entirely on whether that model actually exists. The same batch of Jev repos appearing together warrants waiting and watching. #LLMs #AI Agents #SDKs #jev

Native vLLM serving for Jev decision models

mode-ioPython★26
C x-ladon-x-argus-x-gorgon-x-khronos A script that batch-generates TikTok risk-control signature parameters (x-ladon, x-argus, etc.), with some genuine reverse-engineering content, but its purpose is bypassing platform anti-scraping, making it an abuse tool. Research value doesn't match the risk. #Reverse Engineering #Social #tiktok

tiktok api algorithms generate x-ladon, x-argus, x-gorgon, x-khronos

pingwin010Python★52
C dsh-our-free-model Claims a plugin gives unlimited free access to several frontier models, without saying where the quota comes from, whose keys are used, or where data goes. Such free-ride entry points usually rely on private proxies or someone else's credentials, can fail anytime, and carry account and privacy risks. #LLMs #Crack #Scams #Plugins

All you do is install this plugin in dsh: no login, no sign-up, no API key — the frontier models are just there, Muse Spark 1.3 and MiMo V2.6 among them. Completely free, with no usage cap.

zouyuxuan122JavaScript★90
C jev-gui-delegate A wrapper layer for Codex to operate Windows and Chrome GUIs, built on task contracts plus semantic decisions. But it appears in bulk under the same Jev name, heavy on jargon and light on verifiable content, looking like part of the same promotional wave. #AI Agents #Automation #Windows #jev

AI-assisted Windows and Chrome GUI delegation for Codex: task contracts, local execution, Jev semantic decisions, recovery and outcome verification.

YUTA-fywooPython★21
C sample-voice-ai-agent-studio An AWS official sample-style voice agent workbench whose repo description is even blank, apparently an internal demo copied straight over. Unless you happen to be choosing a stack on AWS voice, there is nothing to take away. #AI Agents #Speech #SDKs

aws-samplesPython★19

DSkip46

D System-Design-KIT Aggregates system design resources, case studies, and roadmaps to help beginners build architecture knowledge; 20 stars in 0.5 days is routine list-type exposure. It could serve as a course funnel, but awesome lists are highly homogeneous with high maintenance costs and hard to turn into a paid standard dependency, so both technical and commercial differentiation are weak. #Learning

A curated collection of resources, case studies, tools, and roadmaps to help you go from system design beginner to confidently architecting large-scale distributed systems.

avinash201199Language unknown★20
D leadsheet.fm An atproto-based guitar tab and chord sharing app addressing sheet-music collection and social spread; 19 stars in 1.1 days is modest. It could monetize via subscriptions or catalog partnerships, but atproto music apps have small user bases and lag far behind giants like Ultimate Guitar, so skip. #Social #Web Frontend

A guitar chord sharing app built on atproto

jazwareTypeScript★19
D recipe-lab-sony-pmca-cn Localizes Sony's camera film-recipe app into Chinese and supports writing recipes to the camera, meeting a real localization need; 38 stars in 0.9 days comes from translation arbitrage, not originality. Hard to commercialize independently and constrained by the original author's authorization and PlayMemories platform retirement, so skip. #Images #Android

Chinese localization of the Sony camera film recipe app Recipe Lab: 77 film color recipes, real-time preview and write to camera, effective long-term in all shooting modes. Compatible with A7R II / A7…

jcx396905-gifJava★38
D DiscordFix-Discord An unofficial Discord Windows repackage with connection configs, apparently meant to bypass network restrictions; 83 stars in 0.1 days is abnormally fast, looking like community reposting or star farming. It violates platform terms with no compliant commercialization path and could be taken down at any time, so skip it. #Windows

Updated DiscordFix build for Windows with Discord connection settings and refined configurations.

Helixdechimney62Batchfile★83
D beebots Runs three model-driven trading swarms on OKX with paper trading by default, an agent+crypto experiment; 18 stars in 0.4 days is modest momentum. Monetization depends on selling strategies or subscriptions, but profitability is unproven, crypto compliance risk is high, and trading bots are crowded, so skip. #Crypto #AI Agents

Three AI trading bees on OKX, every decision by Jev. Paper trading by default. Not financial advice.

imikerussellTypeScript★18
D Syynapse-respocket-executor Provides a lightweight Lua execution and debugging environment for Roblox-like scripts, solving fast script iteration; 20 stars in 1.7 days shows initial heat but mature executors of the same kind already hold the space. Monetization can only be paid closed source or gray distribution, with high compliance risk and no chance of becoming a standard dependency, competing against long-standing commercial executors. #Game Cheats #Automation

Synapse is a lightweight Lua scripting environment designed for script development, testing, and debugging. It offers a rich set of features, stable performance, and an intuitive interface, making it …

0wycieLua★20
D captains-deck A read-only plugin for the Herdr dashboard with a single function, fully dependent on one host product; 26 stars in 1.2 days is okay but the audience is extremely narrow. It can only supplement the plugin ecosystem and can't commercialize independently, becoming useless if the host product isn't popular, so skip it. #Productivity #Plugins

Captain's Deck - a read-only Firstmate flow kanban plugin for Herdr

deimantasnorkPython★26
D seedance2.0mini-seedance-2.0-mini-api-cn A seedance-2.0-mini resale gateway pushing low-price top-ups and an OpenAI-compatible interface; 21 stars in 0.8 days comes from SEO keyword stuffing rather than real users. It has no technical moat, razor-thin margins under price wars, and may violate model vendors' resale terms, so skip it. #APIs #Video

seedance-2.0-mini / seedance2.0mini API: 480P-input $0.0064; 480P $0.0106; 720P-input $0.0138. Model IDs, parameters, curl and Python examples, OpenAI-compatible interface, top up from $1.

devbort2lqizwjShell★21
D seedance2.0mini-seedance-2.0-mini-api Yet another seedance-2.0-mini resale gateway, acquiring users with per-second pricing and a compatible interface; 21 stars in 1.4 days and highly isomorphic with other repos in this batch, clearly copy-and-spam. It has no differentiation or technical moat and its business is unsustainable, so skip it. #APIs #Video

seedance-2.0-mini API (seedance2.0mini / seedance 2.0 mini): 480P $0.01056 per second; 720P $0.02288. Model id, api settings, curl and Python examples over an OpenAI-compatible gat

hhad58x5d3nht3aShell★21
D seedance2.5-seedance-2.5-api A seedance-2.5 resale gateway with fixed per-resolution pricing, one of the batch's spam accounts; 20 stars in 1.4 days comes from keywords rather than product strength. Margins are squeezed by upstream pricing, official and aggregation platforms are strong substitutes, and multi-account operation itself is a compliance risk, so skip it. #APIs #Video

seedance-2.5 API (seedance2.5 / seedance 2.5): from $0.0961 per second (480P); 720P $0.216; 1080P $0.3849. Model id, api settings, curl and Python examples over an OpenAI-compatibl

hvk5d6snsba15uumShell★20
D MuseAI-Skills An unofficial archive of muse.ai skill and runtime files that produces no capability itself, just reposting a 68-skill list from elsewhere. Slightly useful for studying its harness structure, but it's neither author nor spec, may drift from upstream anytime, and skill collections like this are already plentiful. #Agent Skills #AI Agents #MCP #AI

muse.ai (Muse AI) skills and runtime snapshot: 68 skills, workflow guides, connector manifests. muse.ai skills and runtime files; unofficial archive / Unofficial archive.

win4rHTML★125
D seedance2.5-seedance-2.5-gateway Essentially a resale gateway for seedance-2.5, attracting developers with low prices and one-click keys; 21 stars in 1.4 days comes from keyword stuffing rather than user word of mouth. Profit depends on upstream price spreads and is easily replaced by official APIs and aggregators like OpenRouter, with no proprietary technical moat, so skip it. #APIs #Video

seedance-2.5 API (seedance2.5 / seedance 2.5): from $0.0961 per second (480P) through one gateway key. Model id, gateway settings, curl and Python examples over an OpenAI-compatibl

d3qqcegqn5bf1acoShell★21
D qwenimage3.0-qwen-image-3.0-api-ja A qwen-image-3.0 API relay for Japanese developers, acquiring users through keywords and a low minimum top-up; 19 stars in 1.2 days is weak momentum. As a resale channel it lacks proprietary capability, with official direct access and aggregation platforms as strong substitutes, and its business is fragile to upstream pricing, so skip it. #APIs #Image Generation

AI API relay/gateway Japanese guide · qwen-image-3.0 / qwenimage3.0 API: default $0.0206; 1K $0.0206; 2K $0.0206. Model IDs, parameters, curl and Python examples, OpenAI-compatible interface, top up f…

fpbjogs5demjfShell★19
D gptimage2.5-gpt-image-2.5-4k A 4K resale gateway for gpt-image-2.5, pricing by resolution to attract developers; 19 stars in 1.4 days with little community interaction. It has no proprietary model or service capability, is easily replaced by official APIs and aggregation platforms, and has razor-thin margins under price competition, so skip it. #APIs #Image Generation

gpt-image-2.5-ext API (gptimage2.5 / gpt image 2.5 ext): 4K $0.021 per image; 2K $0.014; 1K $0.0085. Model id, 4k settings, curl and Python examples over an OpenAI-compatible gatew

gkdh0uepxud4bShell★19
D respocket-fh-trainer-mod-menu A cheat trainer for Forza Horizon 6 meeting the demand for cheating; 20 stars in 0.7 days is average momentum. It could only monetize through private distribution or paid cheats, directly violating the EULA with ban and takedown risk, and it must compete with established cheat studios, leaving no compliant commercial or standard dependency space. #Game Cheats

Advanced trainer for FH6 or Forza Horizon 6

ajisaid2C++★20
D ssd-tweaker-suite A dual-platform suite claiming to optimize SSDs, control TRIM, and extend lifespan, full of emoji and "maximize IOPS" talk, hitting 101 stars in 0.1 days. System-level disk tools appearing with this kind of packaging carry far more risk than benefit; avoid. #Performance #macOS #Windows #Hardware

⚡ High-performance SSD optimization & TRIM control suite for macOS & Windows. Reduce write cycles, maximize IOPS, enable TRIM, and extend drive lifespan.

lavenderrescuerblazeLanguage unknown★101
D jev-pokemon No description; from the name it looks like a small Pokemon-themed project, again carrying the Jev cluster tag. With no gameplay explanation or code structure to inspect, it reads more like a topical toy. #Games #LLMs #jev

christianmatTypeScript★31
D ag-login-inject A tool that bulk-loads Google accounts into a routing service to freeload model quota, bypassing quotas and abusing accounts. Engineering-wise unoriginal, compliance-wise problematic. #Automation #Scams #LLMs

Bulk-provision Google accounts into 9router for Antigravity AI model access (ag/claude-*, ag/gemini-*)

rullxdPython★24
D homebrew-tools-macOS Just the author's personal Homebrew tap collecting self-use scripts and CLIs, with very limited general value and collaboration; 17 stars in 0.6 days is routine exposure. It has no standardization potential and no monetization path, and is fully homogeneous with the vast number of personal taps, so skip it. #CLI

🍺 Personal Homebrew tap for custom macOS formulas, CLI utilities, and developer scripts.

CaptureGrubEnchantRuby★17
D DiscordFix-Network A Russian-described one-click batch package for fixing Discord networking, appearing in bulk under multiple accounts with the same template, dressed up with emoji and "optimized config" talk. Such packages usually bundle executables of unknown origin; best to avoid entirely. #Networking #Windows #Scams

Updated DiscordFix build for Windows with Discord connection settings and refined configurations.

Seizetenhold47Batchfile★84
D DiscordFix-YouTube Nearly word-for-word identical to the previous entry, just with a different repo-name suffix, a copy from the same template farm. No source notes, no version history, and the only signal is overnight stars; not recommended to download. #Networking #Windows #Scams

Updated DiscordFix build for Windows with Discord connection settings and refined configurations.

ShadowglidevastateBatchfile★82
D tiktok-views-bot A TikTok view-count bot packaged together with its signing algorithm. It's a gray-market platform tool, neither compliant nor offering any design tradeoffs worth learning from. #Social #Automation #Scams

tiktok views bot tool x-ladon x-gorgon x-argus full api tiktok x-khronos, x-ladon, x-argus, tiktok device register, tiktok api

pingwin010Python★46
D CapCut-Pro-macOS-Windows A cracked distribution repo labeled "CapCut Pro unlocked Premium on all platforms," essentially pirated resources plus a downloader with unknown risks. Editors can turn to legitimate free alternatives like DaVinci Resolve; there's no reason to touch it. #Crack #Piracy #Video Editing #macOS

Capcut Pro Unlocked v2026.7 - Full Premium (Windows & MacOS)

SpecterLouseLanguage unknown★177
D seedance2.5-seedance-2.5-prompts Named a prompt repo but actually a funnel page for a video generation API reseller gateway, full of per-second prices, model IDs, and curl samples. It offers no reusable prompt methodology and doesn't say who operates the service, so there's no verifiable value for developers. #Video #Prompting #APIs #seedance

seedance-2.5 API (seedance2.5 / seedance 2.5): every recipe runs from $0.0961 per second (480P). Model id, prompts settings, curl and Python examples over an OpenAI-compatible gate

xfeqaolq2ovnvShell★22
D gptimage2-gpt-image-2-api-ja Yet another price sheet for an image generation API relay, listing model IDs, parameters, minimum top-up amounts, and client examples. It implements nothing and can't confirm the service's origin or stability; a keyword-stuffed repo with no decision value for users. #Image Generation #APIs #Prompting

gpt-image-2 / gptimage2 API: default $0.0085; 1K $0.0085; 2K $0.014. Model ID, parameters, curl and Python examples, OpenAI-compatible interface, top-up from $1.

yd7nkljqb7Shell★22
D 2026_match_model_A_openproject Code and a paper draft for problem A of a graduate mathematical modeling contest, a one-off competition artifact. Possibly useful to fellow contestants, but the problem setting is detached from real scenarios and the code has no reusable interfaces or docs, so as an open-source project there's nothing for others to build on. #Science #Data Analysis #Learning

Open-source solutions for Problem A of the Huawei Cup 2026 Graduate Mathematical Modeling Contest.

zhuangzhuangaiyutingyaPython★16
D gptimage2-gpt-image-2-api-cn An image API relay repo from the same source as the previous entry, a Chinese price list plus examples, almost word-for-word duplicated. No engineering beyond model calls, and service stability and compliance can't be verified; a template-filled placeholder project. #Image Generation #APIs #Prompting

gpt-image-2 / gptimage2 API: default $0.0085; 1K $0.0085; 2K $0.014. Model ID, parameters, curl and Python examples, OpenAI-compatible interface, $1 minimum top-up.

zjk4vwp01tpt07Shell★22
D weather-app A weather forecast app in Kotlin with 26 stars in a day, but nothing beyond a teaching example is visible, and it doesn't state its data source or differentiators. Worth a look for learning Compose, but of no reuse value to outside developers. #Android #Learning #APIs

Kamp-TeachingKotlin★26
D parallax A Swift repo whose README says nothing about what it does, and the source of 35 stars in 1.4 days can't be judged. Until it explains its purpose, it isn't worth clicking into. #macOS #3D #UI Design

bddickenSwift★35
D gptimage2-gpt-image-2-api-ko A Chinese promo page reselling a third-party image model API, just a price table plus curl examples; the repo is only an SEO landing page. The same template appears a dozen-plus times under the same batch of accounts, with no technical content. #Image Generation #APIs #LLMs

gpt-image-2 / gptimage2 API: default $0.0085; 1K $0.0085; 2K $0.014. Model ID, parameters, curl and Python examples, OpenAI-compatible interface, top-up from $1.

jlqhf90669Shell★21
D seedance2.0mini-seedance-2.0-mini-api-ja A Japanese price explanation page for a video generation model API, listing model IDs, parameters, and sample requests, with no code in the repo itself. Such reseller promo repos appear in bulk overnight and aren't worth individual attention. #Video #APIs #LLMs

seedance-2.0-mini / seedance2.0mini API: 480P-input $0.0064; 480P $0.0106; 720P-input $0.0138. Model ID, parameters, curl and Python examples, OpenAI-compatible interface, top-up from $1.

kuhl8wgnph0Shell★22
D gptimage2-gpt-image-2-api Another API reseller repo on the same template as the earlier entries, only the language and price tiers changed. For developers it's advertising rather than a tool; official docs are more trustworthy. #Image Generation #APIs #LLMs

gpt-image-2 API (gptimage2 / gpt image 2): 1K $0.0085 per image; 2K $0.014; 4K $0.021. Model id, api settings, curl and Python examples over an OpenAI-compatible gateway, $1 minimu

kywh6cljfydwShell★20
D Fb-Report A site that mass-reports Facebook accounts via bots, an attack tool abusing the platform's reporting mechanism and directly violating terms of service. Not worth reading the code, and it shouldn't get exposure. #Social #Scams #Automation

Fb-Report es un sitio web programado para denunciar cuentas de Facebook mediante bots automatizados.

lward4Language unknown★35
D kling3-kling-3-gateway A price list plus call examples for a video model API gateway, identical to a dozen repos in the same batch with only the model name swapped. It's channel promotion, not an open-source project. #Video #APIs #LLMs

kling-3 API (kling3 / kling 3): default $0.0672; pro $0.0896; sound $0.1008. Model id, gateway settings, curl and Python examples over an OpenAI-compatible gateway, $1 minimum top-

nm23y450v9vShell★19
D nanobananapro-nano-banana-pro-4k A reseller explanation page for the 4K tier of an image model, with prices and parameter examples. A templated promo repo with no implementation to evaluate. #Image Generation #APIs #LLMs

gemini-3-pro-image-preview API (nanobananapro / gemini 3 pro image preview): 4K $0.04 per image; default $0.03. Model id, 4k settings, curl and Python examples over an OpenAI-compa

nmkshs742wz1yrhShell★20
D aiapi-gateway-pricing A price summary repo for an AI API gateway, ending after a few unit-price tables; even the service itself isn't here. Less useful to developers than just reading the vendor's docs. #APIs #LLMs #Fintech

AI API gateway pricing: published unit prices for image2.5 ($0.0085/image), Seedance 2.0 Mini ($0.0106/sec), LLM tokens from $0.0229/M. One OpenAI-compatible key, $1 minimum top-up

o6vm1o9tShell★21
D seedance2.5-seedance-2.5-api-cn A Chinese price page for a video model API, the same template recurring across multiple accounts. No implementation, no tests, only quotes; not a project worth following. #Video #APIs #LLMs

seedance-2.5 API Chinese documentation (seedance2.5): 480P $0.0961 per second; 720P $0.216; 1080P $0.3849. Model ID, parameters, curl and Python examples, OpenAI-compatible interface, top-up from $1.

ok66oqyea0fi8zqvShell★18
D ai-api-proxy Yet another API gateway price promo repo, duplicating content from the same batch, with even the description stitched together. It's called proxy but contains no proxy implementation. #APIs #LLMs #Networking

AI API gateway pricing: image2.5 $0.0085/image, Seedance 2.0 Mini $0.0106/sec, LLM tokens from $0.0229/M. One OpenAI-compatible key, $1 minimum top-up, per-call cost in every respo

pbd91ed197Shell★19
D claudeopus4.7-claude-opus-4.7-api A price page reselling large model APIs, where even the model version numbers look like placeholders. Same template as a dozen-plus repos in the batch; an SEO farm. #LLMs #APIs

claude-opus-4.7 API (claudeopus4.7 / claude opus 4.7): input $4; cached_input $0.4; cache_write_5m $5. Model id, llm settings, curl and Python examples over an OpenAI-compatible ga

rbvzaxis5myenlShell★18
D nanobananapro-nano-banana-pro-2k-fr A French price explanation for an image model API, again just a channel promo landing page, matching the English version sentence for sentence. No implementation worth evaluating. #Image Generation #APIs #LLMs

nano-banana-pro / nanobananapro API: default $0.03; 4K $0.04. Model ID, parameters, curl and Python examples, OpenAI-compatible interface, top-up from $1.

t7xirxsvznf39Shell★18
D seedance2.5-seedance-2.5-720p-ko A Korean price page for a video model interface, sharing one template with the same batch, only resolution tiers changed. It's reseller promotion, not a project. #Video #APIs #LLMs

seedance-2.5 / seedance2.5 API: 480P-input $0.0576; 480P $0.0961; 720P-input $0.1296. Model ID, parameters, curl and Python examples, OpenAI-compatible interface, top-up from $1.

u04jznqShell★22
D gptimage2.5-gpt-image-2.5-api-fr A French reseller explanation for an image model API, with prices and curl examples. Such repos appear in batches in the same period, clearly promotional material farming exposure. #Image Generation #APIs #LLMs

API gateway/relay — French guide · gpt-image-2.5 / gptimage2.5 API: flare@1K $0.0085; sunburst@1K $0.0085; flare@2K $0.014. Model ID, parameters, curl and Python examples, OpenAI-compatible interface,…

uq4j9o9h9oaoiShell★19
D seedance2.5-seedance-2.5-1080p A 1080P-tier video model API price page, another copy in the same repo series. No code, no benchmarks, only a quote table. #Video #APIs #LLMs

seedance-2.5 API (seedance2.5 / seedance 2.5): 1080P costs $0.3849 per second; 720P $0.216; 480P $0.0961. Model id, 1080p settings, curl and Python examples over an OpenAI-compatib

vsm67u54g4jShell★22
D Pubg-Cheat A PUBG cheat repo with no description or technical notes, just finished executable-style distribution. There's nothing for developers to learn, and using it means account and legal risk; don't look, don't touch. #Game Cheats #Games #Windows

FuturemaccareLanguage unknown★24
D PUBG-External An external cheat for PUBG, no description, no source explanation, distributed as a finished product. Beyond ban and malware risks, such repos offer no value. #Game Cheats #Games #Windows

savagedragonrevive7Language unknown★24
D moonbox-gift A pure HTML repo with no README and not a single sentence saying what it does, yet 24 stars. By the rules its purpose can't be determined, and there's no technical content to evaluate. #E-commerce #Marketing #Social

Yussica1026HTML★24

Show HN 2

Show HN: Agentic CUDA Kernel Optimizer

Show HN

Uses langgraph to orchestrate agents running C++ harnesses and reading nsight results to auto-edit CUDA kernels, automating the manual performance-tuning loop; HN 36 points and 27 GitHub stars corroborate. It could offer optimization services or join MLOps platforms, but results are hard to reproduce and NVIDIA's own toolchain covers this.

Hello; I was working on optimizing some CUDA kernels and I thought may be it is a good oppurtunity learn langgraph as well. I created a simple C++ CUDA Test Harness and handed that to AI agents. They …

#AI Agents #GPU #Performance

▲36

Show HN: Jev Plays Pokémon Red

Show HN

Lets low-latency model Jev play Pokémon Red and streams token costs, testing long-horizon agent planning; 259 points is the batch's strongest signal, with open source plus streaming driving recurring traffic. It could funnel to an agent runtime or model promotion, but entertainment projects rarely become products and it depends heavily on one model's capability.

Hey HN! Wanted to share a fun project I've been hacking on. Given Jev can make decisions really fast (but not fast enough to play Doom yet sadly), I wanted to try and push it to play a more complex ga…

#AI Agents #Games #LLMs

▲259

Hot New Repos

affaan-m/ECC★265957 The agent harness performance optimization system. Skills, instincts, memory, security, and research-first development for Claude Code, Codex, Opencode, Cursor and beyond.
thedotmack/claude-mem★94291 Persistent Context Across Sessions for Every Agent – Captures everything your agent does during sessions, compresses it with AI, and injects relevant context back into future sessions. Works with Claude Code, OpenClaw, Codex, Gemini, Hermes, Copilot, OpenCode + More
paperclipai/paperclip★83205 The open-source app everyone uses to manage agents at work
NationalSecurityAgency/ghidra★79483 Ghidra is a software reverse engineering (SRE) framework
pbakaus/impeccable★71171 The design vocabulary for your coding agent. Steer with polish and bolder, review with critique and audit, catch AI slop before it ships.
rohitg00/ai-engineering-from-scratch★56959 Learn AI Engineering! Learn it. Build it. Ship it for others.
heygen-com/hyperframes★52928 Write HTML. Render video. Built for agents.
anthropics/claude-plugins-official★36626 Official, Anthropic-managed directory of high quality Claude Code Plugins.
debpalash/VoiceStudio★35785 VoiceStudio is the open-source, fully-local ElevenLabs alternative — voice cloning, voice design, video dubbing, dictation, transcription & audiobook creation in 646 languages.
VectifyAI/PageIndex★35730 📑 PageIndex: Document Index for Vectorless, Reasoning-based RAG

A · Worth watching

Show HN 2

Hot New Repos

Star gain 1d