Custom agents

Understanding LLM thinking from the first principle of next-token prediction

AI 驱动的考研备考系统 — 基于本地 LLM + 知识图谱 + RAG，支持数学一、408、英语一、政治全科覆盖，智能出题与遗忘曲线复习

Local-first, offline-capable Agent memory & knowledge service. REST + MCP dual-protocol, hybrid retrieval (vector + BM25/RRF), document/webpage RAG. No LLM required for storage & retrieval

react + nestjs + go 的工作流项目

learning generative models

Lightweight, local-first security testing command-line tool (CLI) built in Rust that scans LLM wrappers, RAG databases, and autonomous AI agents for OWASP Top 10 vulnerabilities

Specialized AI Assistant for Vietnamese legal knowledge extraction and RAG-based document retrieval.

一个基于 RAG 的开源个人知识库，支持文档解析、混合检索、引用溯源和 AI 问答。

AI Study Copilot - upload course materials, get a RAG tutor with cited sources, auto quizzes, mistake analysis and study insights. Next.js 16 + Supabase pgvector + DeepSeek + local embeddings.

An end-to-end clinical data pipeline mapping FHIR to OMOP CDM v5.4. Features DuckDB, RAG-driven semantic mapping, Human-in-the-Loop governance, and OHDSI DQD validation.

Open benchmark comparing web-search engines (Tavily, Exa, Firecrawl) vs LLM-native web search on quality (LLM-as-judge), cost, and latency. Reproducible, fully logged, one-file adapters.

Cross platform AI Linktree mobile app built with React Native Expo, enabling users to create smart AI-powered profile pages with chat, links, content sharing, and personal assistant experiences.

RAG retrieval-time poisoning detector — demand-recency discrimination, canary injection, provenance attestation; fully offline, SARIF+MD reports

DevTool & Legal RAG benchmark using Matryoshka Representation Learning (MRL) with EmbeddingGemma, Supabase pgvector (halfvec) and local SQLite hydration over 36k Spanish Tax Rulings.

Offline Android research assistant with local inference, cited retrieval, and reproducible corpus tooling

Hybrid semantic and keyword search over Oh My Pi session transcripts

Self-hosted AI orchestration with policy, approvals, PII controls, audit evidence, retrieval, and governed MCP tools

Open, independent benchmark of web search APIs for coding agents: Exa, Parallel, Perplexity, Firecrawl, Tavily, Linkup, Brave, You, TinyFish. Scored on grounded task completion against held-out enterprise docs tickets. Search-only and search+fetch boards, model held constant.

Semantic search that shows only what the evidence supports. Often one or two results, sometimes none. Local, training-free.

Self-hosted knowledge service for AI agents: an LLM-maintained wiki and a business graph, searched in milliseconds, far faster than RAG, with no LLM or database on the read path. HTTP and MCP.

RAG for Fiori apps and metadata

14 named silent-failure modes — validation that passes while nothing is being checked. Each with a runnable reproduction, a fixing recipe and a negative control, plus a self-testing linter.

Local-first personal memory and reasoning system built on Markdown, with evidence-aware retrieval, personal analysis, and profile evolution.

Self-hosted MinerU to WeKnora pipeline for parent, child, and raw question-bank retrieval, with read-only ChatGPT MCP access

Multi-task LegalBERT + RAG over Indian Supreme Court judgments: outcome prediction, bias detection, argument mining and FAISS-based legal Q&A.

Your optimized MCP gateway — registry + gateway exposing all managed MCP servers through one token-sparing endpoint

Lightweight hardware tuner, profile manager, and local model inference launcher for llama.cpp (`llama-server.exe').

Market copilot — an LLM grounded in real order book, liquidation and funding microstructure, not vibes: a local trading assistant that cannot hallucinate, over the Wickra core, in ten languages.

基于 LangGraph 的企业知识与数据 AI Agent，集成 Hybrid RAG、MCP Tool Calling、Context / Memory、Trace 与离线评测。

Prepare web content with Crawl4AI and ingest it into Amazon Bedrock Knowledge Bases, with an optional scheduled AWS Fargate (CDK) deployment.

A hybrid retrieval engine for retrieval-augmented generation (RAG) that runs even on a phone

Searchable SOLIDWORKS 2026 drawing API database for macros, scripts, and automation

Portable Agent Skills for building, operating, evaluating, and governing LLM-Wiki knowledge systems.

Production-grade AI platform for automating logistics document processing with OCR, LLMs, validation workflows, and human-in-the-loop review.

AWS Bedrock RAG tutorial: document Q&A on Bedrock Knowledge Bases + S3 Vectors, streamed through CloudFront. Deploy with CDK or Terraform (same stack), Next.js UI, citations, and a diagnostic that proves it streams.

A zero-dependency, drop-in hybrid RAG engine for PHP 8+ with SQLite.

Open-source AI, ranked: apps, agents, models and the tools around them, each with strengths, weaknesses and a score from real GitHub signals. Re-checked twice a day.

Biblioteca de documentação que oferece contexto a LLMs de maneira que consiga criar, de forma autonoma, componentes e circuitos via edição de arquivos do software Hades Simulation Framework.

Visual agent-orchestration canvas for DeepSeek Harness (dsh): compose LLM/MCP nodes into a DAG, run it from chat, and export the board as a DSH preset. Ships its own engine — install and run.

Official command-line client for YouTrack, covering the day-to-day – issues, commands, projects, users — plus the odd jobs: knowledge-base articles, agile boards, admin settings, and raw REST calls when nothing else fits. Built for people at a terminal and AI agents driving YouTrack in a pipeline.

AI-assisted automated UI testing agent that hunts for the holes AI-generated apps ship with.

The AI-native game engine — describe your game, get a complete playable Phaser 3 title with GPT Image 2 art, levels, code, QA, and multiplayer.

Cross-platform Laya decision routing skill for Codex, Claude Code, OpenCode, Pi and other Agent Skills compatible agents.

RedNote / 小红书 / Xiaohongshu (XHS) browser automation skills for Codex, Claude Code and OpenClaw: search, verifiable research, publish and interact.

Official Nitrosend Cursor plugin for campaigns, flows, contacts, and analytics through remote MCP.

Agent skill: delegate scoped coding tasks to your local CodeBuddy CLI | 把边界清晰的任务交给本机 CodeBuddy CLI

AI agent web app: generate style-consistent game assets from reference images (Claude + gpt-image / Gemini)

帮被派去买全组午饭的那个人，30 秒完成一次不会有人不满意的点单。基于麦当劳 MCP 的多人点单 Skill。

麦门圆桌 McRoundTable —— 三个AI军师用你的真实优惠券，吵出今晚最优的一顿麦当劳 | 麦当劳程序员创意开发大赛参赛作品 | 基于官方 MCP

An AI chief of staff for Microsoft 365, built on Work IQ. Every send stays one approval away.

融合版 Android DSH：预置 bionic Node 24 + PRoot 双运行时、22 个 android_* 设备工具、离线语音唤醒与桌宠。独立包名 io.dsh.fusion，可与其它 DSH 安卓版并排安装。

FOX-1: an open-source AI agent you wear. A live fox on an Android device, worn as a charm, pin or watch.

Every AI, one keystroke away. ChatGPT, Claude, Gemini and Perplexity side by side, a private local model (llama.cpp, MLX, vLLM) and an agent that browses for you. Alt+Space on macOS, Ctrl+Space on Windows and Linux.

AI travel planner: chat to draft an itinerary, build it day by day with a guide, and see it as a plan, sketchbook, map and 3D route with a live budget.

Claude/Agent Skill: Ozon Seller API + Performance API — full official OpenAPI specs (508 ops) with lookup CLI

基于 Spring Boot 与 LangGraph 的智能点餐 Agent 系统，融合 LLM、工具调用与业务流程编排，实现智能问答与订单服务自动化。

A bounded Riftbound rules engine and AI-agent toolkit: Deck Coach, Rule Consult, and Player2 infrastructure with versioned grammar and re-runnable evidence for covered mechanics. Not a game simulator or official rules authority. Licensed card programs live in a private overlay. Unofficial fan project.

Autonomous headless Ghidra 12 reverse engineering for Pi, omp, and MCP clients: decompile, xrefs, call graphs, strings, memory, and patches without the GUI.

AI 原生 CTF 自主解题智能体平台（ReAct · 多代理 · 状态账本 · HITL 审查）

skill + plugins that let AI agents render HTML interaction pages and collect structured user input from the browser.

把 DeepSeek Harness 装进手机：远程发任务、批审批、盯进度。内置 106 段动画的桌面宠物「大肥鱼」，不连电脑也能在手机上玩。Mobile client for DeepSeek Harness (DSH) with an offline desk pet.

Let AI agents safely understand and operate the exact Keysight ADS context you selected.

Skills, personalities, and recipes for Hermes Agent — drop-in SKILL.md files, ready-made personalities, and migration guides from OpenClaw. Maintained by @mustafaergisi.

MCP server exposing the full 2Captcha API to Claude Code: 43 tools (31 captcha solvers + 5 management + 3 pingback CRUD + 3 webhook event-store + 1 composite solve-and-wait), built on the official 2captcha-python AsyncTwoCaptcha SDK. >=92% branch coverage, mypy --strict clean.

Self-hosted automation runtime for scripts, desktops and AI agents — cron/webhook/DAG scheduling, Node/Python/Desktop executors, MCP server, RBAC and audit logs. 开源的自动化执行运行时：定时脚本、桌面自动化与 AI Agent 共用一张可审计的调度台。

Terminal coding agent that drives chat.deepseek.com or chat.qwen.ai through Playwright — no API key required. Reads/edits files, runs commands, commits to git.

Agent Builder for eve, tested against a live eve runtime, then deployed as a live app to your own GitHub and Vercel.

Minimal Proxmox VE MCP server - one Docker command, read-only by default

Local AI agent runtime in a single Go binary — browser console, sub-agents, memory, scheduled jobs.

Evaluate complete agent systems: capture execution evidence, inspect reports, and compare improvements.

Enterprise-grade, local-first agentic tabular data engine with self-correcting SQL and autonomous GitHub PR auditing.

Герман (Herman) is an AI trading agent built on Hermes and tested for two months at Finam Arena with risk controls, learning loops, and guarded execution.

An unofficial, safety-focused Shodan CLI and universal Agent Skill covering the documented REST, Streaming, Trends, and Exploits APIs. Supports Codex, Claude Code, OpenClaw, and Hermes.

AI-powered marketing team for B2B SaaS — 14 specialist agents, 1 orchestrator, 6 workflow orchestrators. Forked from MatthiasMRC/bmad-marketing-growth with Source Fidelity guardrail.

Agent-native geometry kernel: a Rust B-Rep engine where every operation returns a validity certificate — nine structural invariants, computed under the write lock, failing closed. Outside its verified envelope it refuses with a typed error rather than approximating. The kernel cannot lie.

Unofficial MCP server + async SDK for SberBank Online — browserless auth, reads, and human-approved self-transfers. Direct client↔Sber, secrets never touch the agent.

Portable agent skills for spec-driven Kotlin Multiplatform design and truthful cross-platform verification

Your coding agent has amnesia and tunnel vision. Pensieve is the context engine that fixes both — long-term memory plus a live, queryable view of your running system (code, logs, traces), linked by a graph.

modelese（模型腔）— lexicon, linter & rewrite-gate for the private dialect AI develops on long runs. 新时代文言文检测器：给 AI 黑话打分、打回、翻成人话。CC0 词典。

Task-aware AGENTS.md context injection and compliance reports for Codex, Claude Code, and Antigravity.

Open-source, self-hosted agent platform for banks and fintechs, declarative agents, banking guardrails, CI-gated evals, tamper-evident audit. Beta.

Build AI applications with this curated collection of free LLM APIs, coding copilots, and infrastructure tools.

MCP server and skills for formae: connect Claude Code, Codex, OpenCode, Cursor or any MCP client and manage infrastructure through conversation.

Long-running coding-agent orchestrator for GitHub Projects

Open source AI operating system for any company. An Obsidian vault your AI agent can read, search and act from. Clone it, run /onboard, and it builds a second brain for your business: departments, tools, integrations and memory in one place.

Little bots with personality, for any agent

A memory of the codebase your AI agent can't get from grep: a persistent call graph, semantic search, and per-symbol review learnings over any repo. Self-hosted MCP server — Go, tree-sitter, Apache AGE.

Composable agent skills for natural, context-aware product translation and translation QA.

Local bookmark manager for AI agents

任务架构通用智能体能力包 ·

Agent skill for authentic Arabic writing: strips English calques and translated phraseology (العرنجية) from Arabic text. Works with Claude Code, Codex, Antigravity, VS Code. Based on Ahmad Al-Ghamdi's treatise — مهارة الكتابة العربية الفصيحة ونبذ العرنجية

Text and structure to full-body 3D animation. A Motion Score in, a three.js AnimationClip out: forward and inverse kinematics, balance, ground contact and secondary motion in plain JavaScript, with no mocap data, no model weights and no renderer.

Content safety + governance for AI agents in one import. IBM Granite Guardian risk classification, NVIDIA NemoGuard content moderation, autonomous-send caps, and a tamper-evident audit ledger — over the three.ws Trust Layer.

Minify, measure, preview, and package SVG art for a single v1 transaction.

Visualize how a 4 KiB transaction spans frames and why retransmits matter.

Pay-per-call Papertrade intelligence API for AI agents over x402: liquidation risk, liquidation maps, settlement estimates and top-trader stats, paid in USDC on Solana or Base.

Bind a coin launch to a compact origin story and permanent provenance payload.

Turn JSX into brand images (Open Graph, social posts, banners). No browser, no Figma, no server.

Runs Claude Code in a loop for days. Every iteration is a fresh claude -p, and a commit ships only if your own checks pass.

Hand plan tasks from Claude Code to Gemini 3.8 Flash through the Antigravity CLI, and merge only what passes your tests.