Custom agents

灵犀流序（Lingxi Flow）：运行在 HarmonyOS NEXT 上的 AI 智能体（Agent）个人助理。一句话描述需求，AI 自主规划、调用系统工具完成任务。纯 ArkTS 实现，支持 25+ 系统工具（文件/提醒/日历/天气/搜索/记忆）、主流大模型接入（DeepSeek/通义/Kimi/智谱）、JSON 技能包自由增删、RAG 知识库、语音交互、多设备协同。2026 中国高校计算机大赛（鸿蒙赛道）参赛作品。

High-performance, deadlock-immune Pure Go multi-agent swarm engine for Antigravity Telegram Bots. Features 1200ms streaming throttler, Inactivity Turn Watchdog, Buffer Salvage, and 429 Cold Safe Parking.

Collections of recursive self-improvement (RSI): papers where the loop that improves a system also changes the thing doing the improving.

A typed language for LLM agent systems on BEAM: effect-row tracking, auditable tool effects, typed actors, and OTP supervision, queryable by agents over MCP.

A story companion for Honkai: Star Rail — live subtitle OCR, character chat, and alternate story branches.

A curated list of AI/ML systems powering modern marketing technology — causal experimentation, user intelligence, ad systems, growth engines, conversational agents, and the emerging Agent-to-Agent marketing paradigm.

High-speed dual-process browser agent combining sub-150ms Jev reflex decisions with Stagehand stealth automation and Gemini/Claude System 2 fallback.

START HERE — Anton Dziatkovskii, one page of proof

Real Windows UI automation skill for Codex and DeepSeek. Mouse, keyboard and API level control, occlusion rescue, hard safety gate. Every screenshot is read out of band and deleted at once, so the main session stays clean and light - no image bytes, no stale screens, no 413.

A desktop GUI client for the pi agent (pi coding agent). Run an AI agent team — MCP, skills, memory, IM channels. macOS / Windows / Linux, Electron + React.

One repo, one skill: a self-improving experience layer for AI agents - multi-dataset session retrieval, on-device triage, token compression, smart-mode routing for hard tasks, video generation, and a secured capability harness.

An opinionated loop based automations written in rust

Event-driven agentic development. Webhook in, coding agent out. No polling loops.

Personal AI with her personality trained into the weights (LoRA merged into Qwen3.8-27B). Runs on local hardware, a Mac plus a PC's GPU via llama.cpp RPC, with no cloud model or AI API. Talks, sees photos, takes voice from an Android app, remembers past chats (RAG) and uses tools.

A trustworthy document-grounded RAG chatbot for solar and energy technical support, with manual upload, provenance, citations, fault-code lookup, and hallucination-aware answers.

A type-theoretic, production-grade retrieval (RAG) engine in modern C++ — hybrid BM25+HNSW, GraphRAG, RAPTOR, rerankers, and a conformant RCP/1 protocol server.

DQL knowledge base + RAG pipeline + GitHub Copilot instructions for accurate Dynatrace Query Language generation + MCP server + Copilot instructions for accurate Dynatrace Query Language generation

MCP search and evidence tool for AI agents with query rewriting, source zoom-in, sourced answers, and runtime metrics.

Run PI AI AGENTS on Android for free. Step-by-Step installation guide.

Local MCP server and CLI for AI agents: read web pages and public APIs, triage GitHub Actions runs, check HTML pages, and control a browser or Windows apps after you grant access.

The best free social media publishing and scheduling API. SDKs for Python & Node, MCP server for AI agents, 15 platforms.

阶跃星辰 Harness｜让 AI 在桌面真正开工：订阅与 API、多代理工作流、插件、Mini。A desktop AI workspace powered by Step Code.

A local-first coding agent you run yourself: one Python program serving plain web pages, with no build step, no database, no accounts and no tracking.

Bring a named coding agent into Google Meet with durable Codex tasks, live transcripts, voice, and a private sidecar.

把多个大模型装进同一个桌面工作台：六种工作流模式、多智能体协作、向量记忆与请求侧前缀缓存工程。本地优先，API Key 系统级加密，Tauri 2 + React 18 + Rust，AGPL-3.0 开源。Windows / macOS / Linux。

Cursor Skill：开题或做计划前的全网类似技术调研（GitHub 开源为主），结果在对话里交付

Mochi —— 会成长的桌面 AI 伙伴。VRM 3D 角色与 VRMA 动作导入，支持流式对话、语音、记忆与多模型接入。Tauri + React + Three.js × LangGraph。

Standalone Hermes plugin for Linear Agent Sessions and attributed Linear tools

A Cloudflare-powered AI agent. Inspired by Hermes.

The 500 AI Agents Projects is a curated collection of AI agent use cases across various industries. It showcases practical applications and provides links to open-source projects for implementation, illustrating how AI agents are transforming sectors such as healthcare, finance, education, retail, and more.

A production-grade memory system for AI coding agents. Context Capsules, Tiered Retrieval, Contradiction Resolution.

Hermes 通用自组装工作流引擎 全局调度中枢，跨平台一键部署，多领域自适应｜Universal Self‑Assembling Workflow Orchestrator for Hermes Agent. Global priority Lv.110 strategy‑driven engine, cross‑platform, multi‑domain adaptive, one‑click install.

A Chrome skill for AI agents: let your agent drive your own logged-in Chrome — background tabs, real mouse/keyboard events, and control handed back to you at password, 2FA and CAPTCHA walls. Zero dependencies.

Vexa Agent Control helps engineering teams govern the AI-agent traffic they route through it, with centrally managed policies and verifiable, per-surface enforcement evidence.

百万级 AI Agent 平台架构 — 智能客服 SaaS 实战 (AgentFlow)

用 AI agent 一键生成你的豆瓣读书可视化报告 + 个性化品味洞察。适配 Claude Code、Cursor、Openclaw、hermes等本地 AI agent。

Open-source control plane for coding agents you own — run Codex, Claude Code, Gemini, Cursor and ACP agents side by side with shared history, permissions, handoffs and audit logs.

China macro context for trading agents on Base — x402 micropayments for ADR volatility signals (GDP, CPI, policy, industry data). Four layers: DO·KNOW·NOW·ARK.

BeeCode — free AI coding agent for the terminal. g4f models, no API key, real tools, skills/plugins/MCP, EN+RU UI.

Local-first, AI-augmented CISSP companion with a knowledge base that grows with you (resources → kb → wiki). Webapp + 11 Claude Code skills. Two modes: exam-prep tutor for candidates, senior security counsel for certified pros.

Developer tools, docs, and examples for Skales - the local AI Desktop Agent. Agent Skills, API reference, MCP servers, provider guides.

Autonomous perp-trading agents on Hyperliquid with hard risk limits enforced in code, a tamper-evident public journal, paper + live modes, and an MCP server.

基于微服务架构的虚拟路径MES系统（Virtual Path MES），支持2000+设备并发连接，实现生产过程的实时监控、质量追溯和智能预测。

易语言调试中转 MCP：本地 E5.95 IDE 的读取、可靠修改、编译与运行观测，通过 MCP 交给 AI 调用

Агент email-маркетолога для нескольких проектов сразу: помнит контекст каждого продукта, собирает письма, проверяет перед отправкой и разбирает результат. Работает через MCP-коннектор сервиса рассылок.

Local web control room for oh-my-pi (OMP): run, watch and review coding-agent sessions across all your projects from one browser tab.

Lightweight rust based coding agent, inspired by pi

Universal persona distillation skill for OpenPersona — distill any person or character into a runnable skill pack

The open-source, local-first coding assistant and multi-agent IDE for macOS, Windows & Linux.

A self-improving personal state system: an AI agent reads your journal through PAD emotion, security/hope (G) and growth (XP) lenses, asks for your feedback, and turns it into rules. Python stdlib only.

McDonald's perks hub built on the official McDonald's MCP. Introduces a Coupon Availability Window model that tells you when a coupon is actually usable today and how long is left, ranking everything P0-P3.

Claude-inspired subagent orchestration for Pi coding agent: Fresh/Fork/parallel dispatch, continuation, verification, and Pi-native lifecycle

A local-first AI coding agent and native macOS workspace with durable request queues, live progress, guarded tool execution, and reviewable run history. Powered by LUMI, BroccoliDB, and BroccoliQueue.

DSH 插件（上游 NanmiCoder/dsh-agent-teams 的补充版）：AgentTeams 多智能体团队协作 —— 自然语言组队、任务依赖 DAG、信箱通信、右侧栏树状监测；随包自带 17 厂商 × 10 岗位头像与 17 个厂商徽标（286 个文件），支持自定义美术目录与中文岗位名。

Massive autonomous web research for AI coding agents. ⭐ Free Firecrawl alternative. No API keys. Open source. Sponsor ❤️

TradingView Desktop AI Agent for opencode.ai. MCP server + skills + launchers.

元气方程-Energsolve — Agent竞品分析系统

Neovim config + MCP Python bridge — a code editor built for AI agents

MNC-style pharmaceutical pipeline asset assessment Agent Skill

AI Agent 技能：通过 Agnes AI 免费 API 生成图像与视频（文生图 / 图生图 / 文生视频 / 关键帧动画），支持 Claude Code 等客户端

夸克/百度网盘资源整理 Agent Skill + 本地文案助手 | Agent skill to save & organize your own Quark/Baidu Pan shares, with offline copywriting

麦麦点餐最优解｜WorkBuddy Skill + 麦当劳 MCP + 本地 Python 组合求解。按预算、热量和蛋白质筛选搭配，附无需 MCP Token 的离线演示。

CommerceAgent：面向电商场景的智能运营工作台，支持商品、订单、客户与自动化协同。

Interface gráfica desktop para o prime-agent: streaming, ferramentas, árvore de subagentes RLM, explorador de arquivos e execução por SSH.

MCP-style TCP bridge for Autodesk MotionBuilder, enabling AI agents to inspect scenes, rigs, poses, animation curves, takes, and FPS animation workflows.

Dreaming plugin for OpenClaw memory-lancedb — Light/REM/Deep sleep phases with narrative generation and memory promotion.

微信公众号文章抓取 → Markdown / 纯文本 / JSON。单文件零依赖 TypeScript Skill，Node 直接跑，无需浏览器或 npm install。WeChat official-account (mp.weixin.qq.com) article scraper: single-file, zero-dependency, agent-skill ready.

Free, open-source AI add-in for Microsoft Excel with bring-your-own-model support, local Ollama workflows, AI functions, and reviewable workbook edits.

Local-first AI workspace that turns research into content — collect, organize, ideate, write, generate media and automate.

交大微澜项目展示页：校园无人配送智能体 Demo、产品说明与演示录屏

中文单页简历 Skill：简历衣柜 → 拆解 JD → 匹配打分 → 投递前质检，含浏览器内可编辑 HTML 与导出脚本

雷神 Raijin — the terminal-first Agentic Development Environment. Built in Rust on our own GPU UI framework (Inazuma). World-first: per-command GPU block grids, native Nushell structured output, and TUI-aware block rendering. 97 context chips and 715 CLI completions built into the binary. Zero plugins. Zero compromises.

让 AI 用中国地方方言回复。支持东北话、四川话、粤语等方言，适用于 20+ 个 AI Agent 平台。

prompt to get tasteful and clean, transparent motion graphics

📧 One inbox. Every interface. mimux connects to the mailboxes you already have and serves them out however the job needs: a fast, keyboard-driven web app for you — free, forever — and a REST API, MCP server, CLI and webhooks for your scripts, pipelines and AI agents. All of it on your hardware, none of it through a SaaS.

MCP server for AI-driven food delivery — search restaurants and manage orders on Uber Eats & Thuisbezorgd via 11 MCP tools. Reverse-engineered APIs, encrypted local credentials.

An ultra-clean, minimalist full-stack Next.js app that transforms casual text into metrics-driven, ATS-optimized resume bullets. Powered by Gemini 3 Flash and MongoDB Atlas, it operates 100% free. Features a side-by-side editing canvas, elegant rate-limit degradation, and seamless Tailwind v4 print-to-PDF styles for instant downloads.

Claude in Chrome for any AI you already use: your terminal AI (Claude Code, Codex, any MCP client) drives your real, logged-in Chrome from a side panel, CLI or MCP. No login, no API key, MIT.

Speak 3D worlds into being from any AI agent. compose_scene turns one sentence into a placed diorama plan (mood, palette, and single-object forge prompts) over the live three.ws pipeline; build_world runs the whole pipeline server-side in one call; export_scene merges a forged diorama into one glTF binary with selectable nodes; get_scene and list_s

AI Agent that auto-generates professional Notion templates(workspaces) from natural language.

MagicTeX — a LaTeX editor for AI agents (Claude Code): live PDF preview, PDF comments that drive edits, Visual (WYSIWYG) mode, and git-tracked history. No local TeX install.

Model-driven context-window rollover bundle for the DeepSeek Harness: fresh working context, durable notes, verbatim recent tail, no summarization

An open-source personal agent for macOS that lives in a folder and writes first.

A native Android remote control for OpenCode. Run the agent on your development machine. Control it from your phone.

给 DeepSeek Harness 的三层记忆：常驻事实 + 会话摘要 + 全文检索。注入不花任何模型调用，写回用你当前选的模型。

Local-first Android battery and hardware diagnostics. Import Bug Reports, view local explanations, and optionally use your own AI provider. No account required.

AI-powered local control plane for Docker, processes, ports, projects, and developer environments.

AI agent that actually operates your Windows apps — real clicks & keystrokes, persistent memory, 12+ IM channels. An AI Agent that truly operates your desktop

Desktop pet for WorkBuddy that shows credits and live AI agent status on macOS

Put a personal agent inside your product, at 1/1000 the cost. Personal agent infra that runs VM-less personal agents on Cloudflare: Agent Mail, WeChat, Telegram, cloud browser, long-term memory.

Shared AI agents for teams: one agent per channel, with memory, connected apps and approvals. Open source and self-hostable. Hosted at teamski.in

Private, local-first AI agent: planning, sandboxed tools, four-tier memory, streaming UI and OpenTelemetry — runs entirely on your machine with Ollama.

Reproducible 16GB-VRAM Windows local-agent stack: DSH Desktop + KVMem (llama.cpp fork, KV cache in system RAM) + parameter panel. Flagship: Bonsai 2 CRACK 27B — 100+ tok/s, 262K ctx on RTX 4080. GSQ for intelligence at ~50 tok/s; QQZ default.

Is your backtest real? Deflated Sharpe Ratio, backtest-overfitting (PBO) and lookahead checks with measured false-positive rates. Works with vectorbt, backtesting.py and TradingView.

An LLM inference engine for persistent chat bots and agents. It allows models to forget old and unimportant tokens to keep context small and inference fast.

Agentic vulnerability discovery with proof: autonomous agents hunt bugs, but no claim counts until it is replayed in a fresh container, judged against known bugs, and re-attacked after patching. ZCode (Z.ai/GLM) and Moonshot Kimi Code port of rust-in-peace. PoC or it didn't happen.

A tool-calling agent harness that keeps working when tools fail. Retries, timeouts, loop detection, and honest error reporting, benchmarked against injected failures.

compact 150M typed-decision engine: calibrated probabilities, abstain flag, Jev-compatible /v1/systemone API, local ONNX on CPU. 0.3.0 (68M) beat Jev on typed-decisions.

Tool-using agents judge a failing source useless, yet keep querying it. Code and data for the paper.

A smart job radar that scores roles against a user's resume