Custom agents

Agentic Lab — a hands-on catalog of agentic-system patterns and demos (learn + portfolio). Live: https://cuppibla.github.io/agentic-lab/

Hades — runtime agentic local-first com multi-provider, Claude/Codex/OmniRoute/xAI adapters, browser/computer use, Company OS/Growth sandbox, builders, MCP/plugins/skills lifecycle, approvals e hardening enterprise

Matched Outcomes, Divergent Gaze: How Foveated MLLMs Search Compared to Humans

Your computer finally has a Mind Computing out loud.

Open-source AI auto-reply assistant for your own Telegram DMs. One Python file, runs 24/7 on Termux (Telethon + Groq).

Hemi is a desktop AI assistant for macOS and Windows, featuring voice and text interaction, automatic model routing and cost control, New API sign-in, app-level permissions, and automatic memory.

Peerly Spa & Beauty: French spa website with WhatsApp booking and an AI assistant (Claude). Plain HTML/CSS/JS, Lighthouse 97-100.

Open-source, self-hosted proactive AI assistant — an any-model alternative to Claude Duet. Morning briefs, check-ins, background tasks, approvals before anything leaves.

🎬 Autonomous AI Video Editor Skill for Instagram Reels & TikTok for Antigravity, Claude Code, Cursor & Codex.

Super search is all you need: one query fans out across 9 providers (web, Wikipedia, 7 preprint servers) with a receipt per provider. MCP server + diagnostics webapp; Brave Goggles lenses, paper cards, authorized breaker chain.

CI for tool-using agents: deterministic eval suites with record/replay, contracts, budgets, and regression merge gates.

Open-source sports video analysis and basketball analytics for Claude Code, Codex and AI agents. Video to coaching reports, clips, narration and explained MP4s; player/ball tracking, shot charts, training plans and optional OpenHoopVision, WASB, SAHI and KaliCalib evaluation. Any sport via plugins.

Unattended Xiaohongshu publishing agent: CDP-driven browser automation for a site with no API. 35 posts across 50 days, 20 documented failure modes.

Local-first multimodal CAD agent: photos, sketches & text in — verified, parametric, manufacturing-ready 3D models out (build123d, voxel FEM, topology optimization, MCP). Self-hosted, AGPLv3.

借助微信读书，将你关注的微信公众号自动抓取/总结提炼！手动将待提炼的公众号导入微信读书书架后，一次扫码，每天早上就可以帮你把关注的公众号最新文章读成一页纸，Agent自动取全文 → 双键去重归档 Markdown → agent 生成每日简报。实战运行 24 期 / 733 篇归档 / 16 条真实故障根因。

Build custom AI agents and harnesses with tool middleware, guardrails, and audit for Go, Python, and TypeScript.

QQ 群里的数字生命：人设注入 · 长期记忆 · 表情包按情绪分类 · 自主刷B站+发评论 · 14 页可视化控制台 | DeepSeek Harness plugin

Specton — 고객 요구사항 기반 시스템 기능·데이터 모델 자동 설계 Agent (팀 설계로)

Local-first SQLite synchronization through verified snapshots and configurable row-level merge policies. Zero dependencies, Python 3.10+.

Chat-based task manager powered by Gemini function calling — manage your to-dos in plain language instead of forms.

A software team as Claude Code agent skills — PM, Architect, Project Manager, Sprint Master, Engineer, QA, Reviewer, DevOps, Release Manager. Covers the full SDLC and software quality management flow, coordinated through GitHub issues and committed markdown.

An eagle's nest, literally the high vantage point it watches everything from

Offline, deterministic testing for AI agents: 130 checks on OpenTelemetry traces, A2A and MCP. No LLM judge, CI-ready. — by Dr Srinivas Padmanabhuni, CTO & Co-founder, AiEnsured (testAIng.com)

First open-source implementation of the SafetyDrift framework (arXiv:2603.27148) — predicts when individually safe AI agent actions compound into a safety violation using Markov chain analysis. Works as an MCP middleware.

TradingAgents methodology as a portable, zero-dependency agent skill (A股/股票分析): multi-agent bull/bear debate + 5-tier rating gates + T+5 deferred reflection + A-share data playbook. Apache-2.0.

CoAnalyst: a LangGraph research analyst for OT/IT alert triage with human review.

War-game your hard calls. A council of non-colluding officers plus a mandated dissenter returns the strategy that wins — and what would make it lose. Claude Code plugin.

Fail-closed B2A security gate for autonomous agent actions with signed policies and verifiable receipts.

SWIFT: Synthesizing Workflows via Few-shot Transfer. Amortized LLM agent workflow synthesis from reusable structural priors.

A vector editor you draw with — together with an AI agent.

Build production-grade LLM agents from patterns distilled from real agent codebases.

Agent-driven single-crystal X-ray structure solution and refinement workbench

Chicago Divvy bike-share availability on Google Cloud: a Cloud Run pipeline collects live data every 15 minutes into BigQuery, dbt models and checks it hourly, and a hosted AI agent (Strands + MCP) answers plain-English questions from read-only data tools. Complete code private, only shareable part published in this repo

卡内基梅隆大学 11-768《AI Agents》课程中文详细讲义：覆盖工具使用、长上下文、记忆、规划、编码与计算机使用 Agent，以及 SFT/RL 训练与 Agent 安全

识海 — a memory layer for LLM agents that forgets, consolidates, and grows a personality. Personality is implantable but not a script: a seed is a prior, memory does the rest. 人格不是被设定的, 是被记住的。

Tutorial 1, Fall School on Agentic AI in Science 2026: a small tool-using agent over DarkMatterFM-Registry

AI-assisted simulation pre-processing tool that validates STEP/IGES files for FEA readiness, auto-heals geometry, and generates mesh recommendations

Codemeld agentic editor

Autonomous multi-agent cloud engineering framework mapped 1-1 across AWS, Azure, and Google Cloud Well-Architected Frameworks. Local-first, BYO-LLM, with continual learning guardrails.

The AI-native database. One database for AI agents. Documents, typed tables, graphs, vectors/search, blobs, queues, events and time series in one replicated .NET cluster. Everything links to everything else, and agents reach all of it through SQL, a built-in MCP server and a typed .NET SDK.

An art experiment for simulated continuous inner life: it thinks when no one is asking, remembers across weeks, and becomes a unique entity over time. Self-hosted, single-user, open source.

Agentic Reference Manager and PDF annotation tool

Veistra - an Agentic Development Environment. A desktop shell where technical design and product design tools sit beside the coding agents that build from them.

Agentic audio transcription in Go: fast speech-to-text with parallel chunking, an LLM judge agent that compares and merges candidate transcripts, extensible via Agent Skills.

Horizon: a local-first AI workspace with built-in model loading, tools, artifacts, profiles and agent integrations.

Android calculator suite: 66 calculators in 8 categories, an AI that reads a photograph of a mathematical problem and works it out with the app's own engines, plus a scientific calculator and a full calculus workbench. Easy to use, works offline, no ads, no accounts, no tracking. Kotlin + Jetpack Compose.

Cursora: your AI companion at the cursor. Select text anywhere, press Ctrl+G, ask. Local or cloud models, private by design. Downloads for Windows and Linux.

装在自己设备上的私有化数据工作台｜飞牛 fnOS 原生应用，数据落在自己的 NAS，多端同步（fnOS / Android / Windows / Edge）· Self-hosted personal data workspace on your own NAS: tasks, calendar, timetable, notes, passwords, ledger, team collaboration, AI assistant.

Open-source AI companion for your PC, phone and an ESP32 desk device. Local or cloud brain, real work in your apps, always asks first.

Full-stack e-commerce store with Dr. Eva, an AI skincare consultant (Gemini). React, TypeScript, Supabase, 5 languages. Graduation capstone @ EVA Pharma.

换个模型也不丢故事：给 DeepSeek 大鲸鱼装上本地记忆并支持web可视化管理，帮它记住你们的约定和没做完的事，下次打开还能接着聊。

Wayland popup translator and language-learning assistant with dictionary, vocabulary, and optional AI

A lean browser interface for AI agents.

Nightly memory curation for Claude Code on your Mac. Injects repo and client context at session start and keeps it current while you sleep.

One-line installer for the SentinelX agent — Linux/macOS (curl | bash) or Windows (PowerShell install.ps1)

DSH Search & Proxy: self-hosted SearXNG search + optional process proxy for DeepSeek Harness. Keep official search; the proxy may improve connectivity and downloads. | DSH 搜索与代理：免费自托管搜索＋可选进程代理，继续使用官方搜索时也可用。

Small Anna app for capturing tagged notes, with local storage, Markdown import/export, and a Python tool that returns structured summaries.

Memory for AI agents: beliefs with a source, a certainty and a revision history. SQLite, zero deps, MCP server.

MCP server for TypeSafe AI's Jev: typed yes/no, choice and score answers with act / review / abstain gating.

MCP server exposing StockMarketScan screeners, chart patterns, options flow signals and equities research as tools for Claude Desktop, Cursor, Continue and any other MCP client.

Open taxonomy of Cardano DeFi venues + MCP server: agent rails into Cardano DeFi

让机器人自己改头像、昵称、签名、在线状态，并记住「我是谁」。AstrBot 插件，支持 AI 生图换头像。

Dairy farm management with animal records, veterinary care, vaccination tracking, lifetime reports, milk production and sales, feed, and payroll. AI assistants provide insights and propose actions for human approval, alongside camera-event monitoring.

Hyperbolic Gateway: a self-hosted web browser for agents and humans - a real Chromium browser your LLM can drive. Local-first, no cloud, no API keys, no telemetry. A DIY alternative to browser-use, Playwright agents and hosted browser clouds.

Instagram MCP server for Claude Code and AI agents. Posts, insights, comments, DMs, publishing, hashtag research and competitor analysis across three tiers of access.

Tollgate — the governance gate for AI agents that move money: deterministic policy enforcement, approvals, budget escrow, tamper-evident audit. 零依赖 MCP 治理网关

Hermes Agent plugin: web_extract fallback chain Jina -> Bright Data -> Firecrawl -> Keenable (Tavily optional), checks each result against a reference copy

LLM, Local LLM Manager, a TUI for setting up LLM service

Project-aware pooling proxy for Serena MCP servers: warm per-project instances with leases and an idle-TTL reaper (Windows-first)

Cross-service code graph engine. 19 language parsers, 13 cross-graph resolvers, .gmap binary store, CLI + Python wheel + MCP wrapper. Worker-protective licensing.

Fast PDF tools for AI agents: 31 tools in one Rust binary. CLI, MCP server and library.

Open WebUI tool: let your local AI search and read your offline Kiwix library (Wikipedia, Stack Exchange, iFixit, MedlinePlus...) and cite every source. No internet needed.

Open-source registry of MCP servers – structured metadata, tools, resources and install configs for the Model Context Protocol ecosystem.

Margo is a lightning-fast, Hugo-inspired presentation engine that compiles pure Markdown into beautifully styled, highly scriptable slide decks built for developers and AI agents.

MCP server for Foundry VTT — 135 tools to prep, run and stage your games from any AI assistant. Rust, single binary, no module required.

Skills for AI agents: SKILL.md folders sealed as git commits, with notes and rollback. SQLite, zero deps, MCP server.

NEXUS Web LLM is a browser-based local inference app for compatible GGUF and LittleBit models, featuring custom WebGPU acceleration, metadata-driven configuration, 100 model profiles, automatic memory planning, speed tuning, and runtime diagnostics.

Agent 原生的量化研究与回测框架 | The agent-native quant research & backtest framework for A-shares & futures

一个 agent skill（任何 AI agent 都能用）：把你给的 Instagram Reel 链接自动下载、翻译文案、转载到 B 站（含上传后验真）。你给链接它就办——不抓取、不筛选、不去重。

Free AI agentic harness — one install seats a panel of AI agents on your model (BYOK or free local model). npm install -g agentqon

Policy-controlled MCP server giving AI agents access to Proton Mail through Proton Mail Bridge: Allow/Ask/Deny per action, owner approvals, audit journal.

Hermes Agent ops: 35 modules for running a self-hosted LLM agent 24/7 — cron reliability, cost guards, context compaction, probes, skill scanners.

A safety-first Claude/agent skill for PentAGI — autonomous AI penetration testing with hard legal scope rules, lab-first workflow and install checklists.

Run a free, persistent Telegram AI agent using Google Colab + Google Drive — zero server cost, powered by Hermes Agent (Nous Research) + OpenRouter free models

Agent skill: a general testing discipline — a requirement is covered only when a test shows that violating it fails. Business rules, data invariants, persistence, concurrency, API contracts, security.

When is enough evidence enough? Sufficiency judgments and stopping rules in multi-step retrieval agents. Technical report, code and full logs.

An AI agent for social media management — listening, creator discovery, multi-platform publishing, and trend research across X, Instagram, TikTok, Reddit, and YouTube

An agentic harness for accelerator kernel generation and optimization, with reusable skills extracted from expert implementations.

Open Agent Engine for long-running agents. More agents per GPU. No compaction. Every request rebuilt from a typed state frame so prompts stay within a fixed budget no matter how long the task runs. Built and measured on real software engineering work. AGPL-3.0 or commercial license.

Database and data-system foundations for multi-agent collaboration: a topic guide and paper list organized by the ACDS architecture.

Your pod of AI agents, in one chat.

Architecture-level code review for any coding agent — Claude Code, Codex, Cursor, Gemini CLI and 100+ more. One verdict, every blocker demonstrated by running the code.

Open-source 3D LAN real-time strategy game where LLMs play. Battle local or API models, or let AI Pilot run your civilization and take over anytime. Bring your own model.

Research prototype for durable agent execution facts, deterministic acceptance, crash-state inspection, and immutable skill versioning.

Curated list of self-evolving recommender systems: LLM-agent loops that change recommender config, code, semantic IDs or policy, with the evidence behind each gain. Companion to the survey + audit of 27 systems.

Regression tests for LLM agents: replay a frozen conversation through today's code, test only the next user line. Stale detection, code-judged checks, guard attribution, pass^k and significance tests, pytest plugin.

KUPAS MASTER (老师傅): Experience engineering for AI agents: tacit knowledge, nine-layer cognitive corpus construction, experience corpora and agent skills. Bilingual technical reports and RAG evaluation.

Local-first stock research: calibrated volatility ranges, transparent signals, BYO data providers — never predicts direction

AEGIS is a protective layer for autonomous engineering—combining spec-first workflows, specialized AI agents, TDD, quality gates, and adversarial reviews to build reliable software.