AI API gateway / relay: 307 models behind one OpenAI-compatible key. image2.5 $0.0085/image, Seedance 2.0 Mini $0.01056/sec, LLM from $0.0228/M tokens. USD settlement, pay-as-you-go, $1 minimum top-up.
AI API gateway / relay: 307 models behind one OpenAI-compatible key. image2.5 $0.0085/image, Seedance 2.0 Mini $0.01056/sec, LLM from $0.0228/M tokens. USD settlement, pay-as-you-go, $1 minimum top-up.
Unofficial Python API and brisk CLI for BRiSK market data: sign in to SBI or Tachibana e-shiten BRiSK with a passkey, record the live feed through BRiSK's own decoder day after day, keep each day in your own private archive, and query your history and OHLCV bars through the cloud API
Prompt builder node for Krea-2 / Flux models in ComfyUI. Pick an art style from 300+ curated styles (filtered by category) and add camera settings – model, focal length, aperture, film stock, shot type, lighting. Outputs a natural language prompt. Easily extendable via JSON files.
Cloud MCP gateway that connects a StackChan desk robot to claude.ai — OAuth 2.1 bridge, reflex daemon, pixel avatar toolkit, and the field notes from making it work
Repo blueprint para Claude Code: no trae la app, trae las instrucciones para construirla en tu máquina paso a paso. Un agente que atiende cada chat de WhatsApp como setter y closer: califica, responde la objeción, agenda y escribe el CRM. 16 archivos por fase, 30 versiones fijadas, compuerta de 23 chequeos. En español.
Local-first MCP memory server for AI agents. Zero dependencies, SQLite on your own disk, 52 MCP tools with Mem0-compatible names. MIT.
—
A curated directory of Hundreds of open-source projects by Yemeni developers —PHP Laravel, Flutter, Python, AI, payments, and Arabic/RTL tooling.
Korgo Bot: persistent AI bots with built-in Orgo cloud computer use.
Durable, traceable project memory for AI coding agents. An evolution of Karpathy's LLM Wiki pattern built for codebases.
Local traffic visibility and privacy controls for coding agents. See what Claude Code, Codex and Copilot send to the model; mask secrets and personal data where the client allows it.
—
Save your GPT-6 Astra tokens. Let Luna do the dirty work.
Mythic C2 MSTeams Communication Channel
GLM-5.3-Flash NVFP4 on 4x NVIDIA DGX Spark: vLLM TP4, DFlash2 adaptive draft, measured FP8 vs NVFP4 vs EXL3 vs SGLang
Replicating Jev with a local LLM
Auto Regis and inject 9Router
—
영문 논문 PDF를 레이아웃 그대로 한글로 번역하는 Claude Code 스킬 (BabelDOC + Claude)
Exodus Fake Balance - wallet UI simulation tool for BTC, ETH, SOL, XRP, and 200+ assets. Customize displayed balances in a controlled demo environment with persistent visual settings and screenshot-ready previews. Designed for testing, demonstrations, and UI prototyping without altering actual blockchain balances or transactions.
Skill que dá super poderes de gestão de produto as pessoas
Operator Toolkit for FilmBench
Unofficial MCP server for T-Bank (Т-Банк, ex-Tinkoff) mobile banking: accounts, cards, transfers, bill pay, grocery ordering, tickets & travel — for Claude Code and any MCP client.
Crow — a local AI agent, not a chat box: 30 tools, MCP, memory, skills, goals, subagents, vision, and image generation and editing with Qwen-Image. Runs on crow-nest or llama.cpp on one GPU, or any model via OpenRouter. Windows and Linux.
互动小说 skill · 你选,它写|选项驱动的互动小说 AI 创作规范,装进 AI 助手就能玩,无需部署。interactive fiction for AI agents, choice-driven storytelling
—
Run MiniMax H3 (video + native stereo audio) locally on Apple Silicon (24GB+ RAM, M1+): ComfyUI workflow, measured cost tables, and the dead ends so you don't repeat them
Find and remove AI provenance signals from files you own: hidden Unicode, token watermarks, visible image marks, and C2PA/EXIF/XMP metadata. Zero-dependency Python 3.10+ core; models, GPUs, and system binaries sit behind opt-in adapters.
—
ROSCon 2026 Workshop - Mastering the Jazzy RMW
Telegram api接码、协议号管理面板
An Autonomous Agent for Slay the Spire. Achieved multiple A20H wins with both Ironclad and Defect.
—
A open source desktop app for building Movies with Minimax H3.
Semantic Conditioning Bridge for MiniMax H3
A Benchmark for Retrieving Papers that Inspire New Research
AI API gateway / relay: 307 models behind one OpenAI-compatible key. image2.5 $0.0085/image, Seedance 2.0 Mini $0.01056/sec, LLM from $0.0228/M tokens. USD settlement, pay-as-you-go, $1 minimum top-up.
自己上传表情包,让 AI 看得懂、也会发 · Stickers your AI can read and send
A python proof of concept tool to assist in your payloads classification against EDR/AV's machine learning models
Autoresearch agents that improve your benchmark: independent re-measurement, enforced budgets, results as pull requests
PySide6 GUI for building OpenWrt on Linux and WSL2
Open-source vulnerability prioritization engine that enriches scanner findings with vulnerability intelligence, identifies urgent risks, and automates alerts and ITSM ticketing.
macOS iMessage skill for Grok Bot — local launchd helper bridge to read, triage, search, and send iMessages
A consistent, fair, benchmark for structured extraction.
The official training code of SingProbe.
The pretraining scaffold behind 'Pretraining a Mini Kimi K3' (free book): six-source HF data mix, benchmark decontamination, tokenizer, WSD training loop, run monitoring. 5B tokens for $252.
Katabump续期脚本
Extract reusable visual systems from reference technical diagrams and generate style-consistent figures.
Self-hosted web panel for the SoftEther VPN Server: the complete JSON-RPC management API, live machine monitoring, and a one-command install.
A language model on a Game Boy Color - pure SM83 assembly
dynavec — serverless hybrid vector database on DynamoDB + Amazon S3 Vectors.
Portable PyTorch-native Microduck simulation and training environment powered by mujoco-torch.
Information-Set MCTS-inspired agent and reference simulation for the Pokémon TCG AI Battle Challenge — belief-based decision policy, deck architecture, and evaluation harness for the Strategy Category submission.
数据平台
Local bilingual probability decisions from context, questions, and candidate answers. Independent research preview inspired by TypeSafe Jev.
A langgraph based workflow with a C++ CUDA harness to optimize CUDA kernels
ComfyUI nodes for MiniMax H3: INT8 X2 VAE and asynchronous NVENC saving. 5-second 1088x1920 video with audio in about 60 s on an RTX 4070 12GB, about 30% faster than the FP16 X2 VAE.
AI API gateway / relay: 307 models behind one OpenAI-compatible key. image2.5 $0.0085/image, Seedance 2.0 Mini $0.01056/sec, LLM from $0.0228/M tokens. USD settlement, pay-as-you-go, $1 minimum top-up.
Quyet: calibrated decision models