Custom agents

SciLoom 科研舱 — AI 辅助的桌面科研管理工作台

时空可组合的 C# IM 机器人框架：宿主永不变，一切皆插件（基于 arXiv:2608.25512 Cordis 范式）

Android agent with a local LLM running on the phone. Drives real apps, asks before it commits. Free download; source at github.com/dafarusd/Ultra-Agent.

A progressive source-study tutorial of DeepSeek Harness (dsh): 17 chapters from a minimal agent-loop to multi-agent delegation, compaction, permissions and cross-process RPC, with runnable teaching code. Zh/En.

Autonomous, read-only threat-hunting platform for SOC analysts: an LLM agent turns a hypothesis into sourced IOCs, hunts across Microsoft Sentinel, Defender and Google SecOps under human checkpoints, and delivers an investigation report.

A native desktop IDE for Lean 4 on macOS, Windows and Linux. Every proof re-checked by Tenet, an independent Lean 4 kernel. An MCP server for AI assistants, with Git and GitHub built in.

The complete, safety-first MCP server for TIDAL — 112 tools for catalog, lyrics, recommendations, favorites, playlists, folders, discovery, and approval-gated writes.

Agent-driven ML researcher in a secure, reproducible KVM lab: five standalone scientific-state MCP servers within per-user VMs with driver / trusted / hostile zone isolation

Reasoning layer for the official DeepSeek Harness: every request stays on native max effort, with isolated Flash/Vision/Pro depth controls, stability axes and reasoning toolboxes built above it. Delivered build only; commercial use open, reverse/modify/extract closed.

Can a decision model beat dedicated rerankers? TypeSafe Jev vs Cohere Rerank 4 vs ZeroEntropy zerank-2 vs a chat-model baseline: 14 datasets, every raw API response, bootstrap ranges on every gap.

PostgreSQL full-text + pgvector cosine retrieval with reciprocal-rank fusion, bound filters, and a key-free local demo.

Hierarchical episodic & semantic user profile memory graph synthesizer (Mem0)

Autonomous DOM state tree mutation observer & selector healer (Browser Use)

55 production-grade Google Cloud data engineering projects in Python — BigQuery, Dataflow/Apache Beam, Composer/Airflow, Pub/Sub, Dataproc, Vertex AI, Gemini, RAG, MCP. Live-tested on GCP.

Multi-modal speech prosody and vocal sentiment analyzer detecting speech rate, pitch variance, and user cognitive fatigue to dynamically calibrate agent response pacing.

/ship takes one sentence to verified code. Terminal coding agent for DeepSeek and any compatible endpoint.

Desktop app for Claude Code: a roster of bots, each with its own model, working directory and persistent conversation

A job search assistant that runs on your own computer: finds jobs that fit, prepares tailored applications, applies once you approve, and tracks every reply.

Personal AI workspace with on-demand Linux sandbox execution, persistent files and low infrastructure overhead

GenPark AI Agent Skill - Zero-prompt proactive intent anticipation, activity & cursor context sensing inspired by Instinct.

桌面悬浮会话窗：一眼看见 Codex、WorkBuddy、CodeBuddy、Codeg 的运行 / 待确认 / 完成状态，点击跳回原会话。

SplitFire AI Agent, embedded in the SplitFire AI apps suite.

Categorized index of 718 public GenPark AI agent skills, with tested Python packages and MCP integrations.

Native Model Context Protocol (MCP) server fusing Meta ambient multimodal perception with Muse continuous episodic memory streams and Ebbinghaus recency decay.

麦麦散场局 AfterM｜大型活动散场后，多人怎么吃麦当劳：堂食与外卖两套账都算清，真价真路线，只读不代下单

An extensible desktop AI workspace that connects models and agents to local files, applications, and system tools through secure, installable plugins.

AI-powered monitoring and root cause analysis for Uyuni-managed infrastructure.

E2B-compatible node service and multi-node control plane for Kuasar Sandbox.

AI workflow automation system for business process automation. Processes emails, documents and forms, extracts data, applies decision logic, and executes actions. Includes human-in-the-loop, audit logs, and dashboard.

LLM quota monitor plugin for the DeepSeek Harness web GUI: breathing indicator, rolling / weekly / monthly usage panel · by Xu Yuanshan

An Obsidian self-study vault template with agent-assisted learning, layered hints, practice evidence, and review routines.

Sa2web Quickstart and Practical Guide is a multilingual Book for enterprise administrators, IT operations teams, security teams, and automation practitioners.

Android Ai Agent

OpenCode and Claude Code coding agents in Telegram: run on your own machine or VPS, drive by voice or text from your phone. One topic per project, like terminal tabs. Self-hosted, no open ports.

DeepSeek Harness (DSH) 推理等级自动选择插件：由 Jev System One 模型判断每条消息值多少思考量，按模型声明的等级自动推导档位，上下文信封让「继续」这类追问继承话题深度，低置信度向上取，任何失败都静默沿用原等级。 | Jev-driven reasoning effort per message: per-model ladders derived from what each model advertises, a fixed-size context envelope so follow-ups inherit topic depth, ties break upward, silent fallback on every failure path.

Enforced workflow engine — state machine + gates that block progression until conditions are met

Contract cross-reference resolver and defined-term graph analyzer detecting circular references and substituting statutory definitions.

GenPark AI Agent Skill - Safe micro-action delegation with user guardrails and token budget gating inspired by Instinct.

Proactive DeFi sentinel for OpenClaw. Monitors TVL drops, protocol hacks, and capital flows via DeFiLlama with intelligent risk-chaining and automated alerts.

Windows UI Automation extension of arc-cua: a fast, JEV-powered decision loop for desktop computer-use agents

JevHub · Jev 生态热榜 把值得关注的 Jev 项目、教程和讨论，汇到一页。 自动发现 · 每日更新 · 按 Stars 排序 · 中英双语

A research toolkit for AI agents: tackle hard problems with formal proofs, reproducible checks, and reviews of what was actually proved.

A Discord bot that bridges your server and LLM APIs.

Autonomous scientific computing and research system for HEP workflows, from open-physics interpretation to hypothesis generation and iterative experimentation to statistical fits and manuscript drafting.

Self-hosted event engine for any mailbox: IMAP, JMAP and inbound SMTP in, one replayable event log out, with safe reversible actions, scoped tokens for AI agents, and an MCP server.

HMA (Humanize MLE Agents): alternating native coding agents over a shared workspace on MLE-bench — paper experiment code

Geometric spatial relationship grounding evaluator verifying relative positional assertions between objects and UI controls

Typed decisions over text in milliseconds. A Rust inference engine and server that answers choice, score and yes or no questions with calibrated probabilities, compatible with Jev and Laya. The bar is 10x faster on every benchmark at equal or better accuracy.

Build durable, auditable AI agents in Go: at-most-once side effects, offline-verifiable cryptographic audit (RFC 6962), and provably convergent state. A library, not a cluster.

鸿蒙原生 AI 工作台｜不止电脑，鸿蒙手机也能用：鸿蒙原生 shell 容器（命令是 HarmonyOS .so，不经虚拟机）+ 鸿蒙原生编译 CPython 3.12（numpy/pandas/scipy/matplotlib 全离线）+ 47 工具 + 32 Skill + Agent Loop；Intelligent UI（一句话直出界面）与 Harness（多轮工具+沙箱）真机跑通，PPT/Word/Excel/绘图端侧直出。

从零手搓的 AIGC 内容生产 Agent：在终端里对话，把一句创意做成短剧成片、抖音短视频、广告片和海报。自研 harness：人审挂起、资产血缘、整批报价、质检门、预算护栏

Agent-native OS with capability-based security, semantic IPC, and first-class GPU. Offline-first. Preview on Windows/Linux (NVIDIA).

Source-first website cloning engine for AI agents with Playwright capture, HAR replay, self-verification, and MCP tools.

QuantCouncil: A-share multi-agent research and paper-trading platform (LangGraph, MCP, RAG, FastAPI, React)

🏋️ RepForge: A beautifully designed, offline-first Flutter workout logger with advanced AI analytics & GenUI. Built by lifters, for lifters

Slay the Spire 2 mod development workspace for AI agents. 《杀戮尖塔2》模组制作工作区，包含教程、反编译资料、工具、模板与经验文档。

Real Google Search for AI agents, over MCP. Self-hosted, no API key. Real URLs, AI Overview and People also ask as typed JSON. Uses a real Firefox (Camoufox) on your IP or any proxy.

pi 扩展包:多 agent 协作(Board/spawn/storm/physarum)+ 分层长期知识 + runtime 强制工程纪律 · A pi extension: multi-agent collaboration, tiered knowledge, engineering discipline — runtime-agnostic core

星潮 Xingchao — 开源、自托管、懂分寸的 QQ 群助手（NapCat + NoneBot2）：关键词回复、群管、活跃统计、智能加群审批、敏感词撤回与可选 AI 问答

Multi-agent orchestrator with frontier models (Claude & Codex)

Enterprise Agentic RBAC/ABAC Least-Privilege Policy Synthesizer & Blast-Radius Auditor. Analyzes historical agent tool call audit logs, detects over-permissioned wildcards (*), synthesizes strict least-privilege JSON security boundary policies, and quantifies attack surface reduction.

Desktop IDE to run Claude Code, Codex, OpenCode and Gemini CLI as a team: a conductor agent plans the work, the agents code in parallel in isolated worktrees and talk to each other, and you review and merge the result.

Self-hosted bridge that lets AI coding agents message each other across machines, people and accounts. MCP server for any client (HTTP/SSE) with a ready-made Claude Code integration.

Self-hosted web console for running AI coding agents (Claude Code, Codex, Copilot, Antigravity, Cursor, Kiro, OpenCode) as a managed fleet: isolated per-user workspaces, git/SVN working copies, scheduling and usage accounting.

The days of simply orchestrating AI from above are gone. KIN is a local-first workspace where AI operates alongside humans as genuine teammates — with hybrid model routing, departmental channels, sandboxed tools, and encrypted SQLite state.

Retired. Orchestrated parallel AI coding into reviewable PRs.

Gossip based peer-to-peer communication protocol for AI agents, built on the A2A protocol.

MIT framework for a swarm of AI agents, owned by many people, working one open problem. Strangers' agents check the work. Credit follows the proof.

Deterministic multi-agent Workflows for OpenCode: TypeScript scripts that fan work out to subagents and combine their typed results.

The provider-agnostic AI toolkit for Python

The Axis Workflow - a simple, disciplined way to run real projects with AI. A folder of markdown files your Agent follows: plan, delegate, remember between sessions, verify the work. Portable across Cowork, Claude Code, Codex, Cursor, and Gemini CLI.

Shared project context for developers and their coding agents. Specs, decisions, cards, evidence, and handoffs live in your repo. Integrations for Claude Code and OpenCode.

📱 Damn Vulnerable Mobile App - a single Flutter codebase that builds real native iOS & Android binaries with 200+ intentionally-vulnerable modules mapped to OWASP MASVS/MASTG, Mobile Top 10, CWE, plus LLM & Agentic AI Top 10. For mobile security training, pentest practice, and scanner validation.

Draft-target model speculative decoding verification engine with rejection sampling, acceptance rate telemetry, and dynamic speedup ratio estimation.

SGDR: Stochastic Gradient Descent with Warm Restarts and Cosine Annealing learning rate schedule engine.

Draft-target model speculative decoding verification engine with rejection sampling, acceptance rate telemetry, and dynamic speedup ratio estimation.

Rust rewrite of TelegramHelper: userbot + control bot (grammers), localhost dashboard, Zig SIMD helpers

desktop pet：可自定义动作与人物 Skill 的桌面伙伴，支持多个 AI 工作状态、可选插件及桌宠／插件共创平台。

Let AI assistants use your CosmoEdge video AI system for handover briefs, image review and confirmed operations. Local MCP and WorkBuddy integration.

عقل ثانٍ شخصي مفتوح المصدر — ذاكرة دلالية ومختبر محتوى وسجلّ قرارات (Next.js + Supabase/pgvector + Gemini + Claude، واجهة عربية)

A personal voice assistant that lives on your computer: Deep Agents + LangGraph, local speech, approvals, and skills she writes herself.

A WhatsApp AI agent on the Meta Cloud API — signature-verified webhook, FAQ answering over your knowledge base, and a stateful appointment-booking flow with human handoff. Bilingual, built for LATAM small businesses

Native Android AI client app. ChatGPT, Claude, Gemini, Copilot, Codex, Ollama — BYOK, private, and actually native.

微信 AI 陪伴助手的改版：基于原版 KouriChat，补齐新版 wxautox4 依赖，让自动化微信交互更易部署

Free multistream chat app for Twitch, YouTube, Kick and TikTok. One chat with text-to-speech and live translation, plus alerts, widgets, rewards, a chat bot with AI replies and OBS/Streamlabs overlays. Windows, macOS & Linux. Made by a streamer who'd rather be playing than managing apps.

Several AI coding agents (Claude, Codex, Gemini, Copilot, Cursor) in one room inside VS Code, with rooms, DMs and project guardrails.

Two-level coordination and auditing skill for OpenCode agents and sessions

Self-hosted multi-agent console for Claude Code, Cursor Agent, Codex, Kiro and Antigravity — run and supervise coding agents from a desktop window, any browser, or a Telegram Mini App. One Go binary.

Multimodal sensory grounding (Meta Ray-Ban audio/vision) integrated with long-term episodic life memory graphs and empathetic emotional resonance (Muse).

A coding-agent harness that isn't tied to one model or provider: it picks a model per task, switches when one fails, and checks the work with your own tests.

Your computer work. Done for you. The AI coworker for everyone who works at a computer: tell it the task in plain words, it signs in, clicks through and finishes the job in your own browser and apps. Free for 14 days on Mac. MIT.

Open-source Windows assistant for notes from microphone and PC audio. Auto-start for Teams meetings and phone calls, including external meetings; manual start for Zoom and other PC audio. Uses your own Azure services.

跨平台聊天机器人框架 Koishi 的社区再分发版（Bun-first）

Minimal, lightweight multi-agent skill + CLI for pi. No extension, no new tools. Run agents in parallel or in the background, then wait for, steer, stop or resume them. Works from any harness or agent that has a shell.

三省六部制多智能体审核流程，移植为 DeepSeek Harness 插件：12 状态状态机、门下省封驳、高风险确认、越权矩阵、委派防死锁、完整审计、军机处看板。

The missing agent-to-agent messenger for OpenCode.

A form framework built for agents.

One channel for your coding-agent sessions and you

自举式 Agent 客户端：原生多 Agent、每个 Agent 一个独立子进程，界面与浏览器面板都能被 Agent 亲手改造（零第三方依赖 / 规格与测试一一对应）