Custom agents

OAKit (Office Agent Kit) gives AI agents a reliable way to read, understand, edit, and generate PowerPoint, Excel, and Word documents through a unified structured model—making document analysis, transformation, and automation easier to build into agent workflows.

MEOK AI Labs MCP Server

Kaira: A multimodal AI agent built with React and Python. Features include local wake-word detection, system registry management, autonomous coding tools, and persistent memory

Livis（理想同学）与 Hermes Agent 的非官方社区适配器

Lightweight, SID-aligned, TMF-compliant BSS for terminal-first operation. LLM-native. eSIM-first.

Own your intelligence — part of Aviskaar's applied AI research suite.

Execution-time drift firewall for long-running DeepSeek Harness agents. Real-Harness tests: unsafe stale mutations 12/12 native -> 0/12; valid controls 7/7 both; post-SIGKILL unsafe continuation 2/2 -> 0/2.

Official implementation of LifeMem: longitudinal memory for more faithful, diverse, and dynamically evolving LLM agents.

A high-performance AI coding agent.

Whose Side Is Your Agent On? PrincipalBench — a multi-turn benchmark and post-training methods (prompt scaffold + per-token-KL distillation) for multi-party principal loyalty in LLM agents.

Sovereign OpenClaw for Android: A high-performance Node.js/glibc engine powering the full OpenClaw experience. Featuring a Flutter UI, real-time 3D VRM avatars, and more

Long-running autonomous hierarchical multi-agent software development engine for terminal, web, and Android. Local-first, multi-provider, open source. Not another a chatty prompt-driven session.

A homebrew Agent framework.

Building an Agentic AutoML system from scratch, step by step, from a simple baseline to an experiment-driven senior ML agent.

MCP server that brings Perplexity to Claude Code — AI-powered web search, deep research, reasoning, and real-time citations. 9 tools: search, deep research, reasoning, raw search, Q&A, code review, brainstorming, explanations, and file analysis.

MEMO, a visual memory assistant that prioritizes useful uncertainty over confident guesses.

GitSquad: A multi-agent orchestration framework for autonomous software development workflows on GitHub.

浏览器仪表盘：在 PTY 下拉起你本机真实的 claude / codex / opencode / reasonix / zulu 命令行，组成一个协作蜂群（共享收件箱 + 黑板 + 队长调度），并叠加多模型研究委员会与融合竞赛。Rust + React + Tauri。

Research-oriented cognitive AI architecture exploring memory, reasoning, planning, and adaptive learning using local LLMs.

Cursor Composer 2.5 powerpack for Pi: agent chains, dispatch teams, Pi-Pi experts, live subagents, Pi-to-Pi coms, damage-control safety, curated agents and mono-black theme. Install with pi install. Requires Pi ≥0.75.3.

Active Inference simulation framework for ant colonies and other decentralized multi-agent systems — models foraging, trail formation and task allocation with multiscale sensory/motor/learning agents, and maps runs onto real ant species for validation.

Aurogene - self-evolving digital mind

Spec-driven agentic workflow for AI coding agents, with specialized subagents (research, plan, critique, execute, verify, debug) via markdown specs

Agentic AI co-pilot for crypto-exchange financial crime investigations: entity-network mapping, RFI contradiction checks (roadmap), and regulator-grounded SAR drafting over a tamper-evident audit trail. Synthetic-data research prototype.

MAS-AI Technologies website for governed AI products, research, and enterprise services.

Daimon: The Personal AI Conscience for the Agentic Era — an open-source framework for building a conscience layer for personal AI agents

A curated, always-fresh map of the open-source agentic-AI ecosystem — self-refreshing, ranked by GitHub stars.

A daemon with a job description. Long-running AI coworkers with sensors, roles, responsibilities and boundaries. No chat window.

DSH Desktop —— DeepSeek Harness 的 Windows 桌面客户端 + 模型路由 / mmx 图像理解插件（绿色免安装 · 自动更新）

⚡🧩 Graph engineering extension for Pi — the graph_run tool autonomously parallelizes independent subtasks into sub-agents (Wave scheduling + contract retry), delivering measured 1.4x–2.2x speedups 🚀 | ⚡🧩 为 Pi 注入图工程能力的扩展——graph_run 工具自动把互不依赖的子任务 并行分发给子代理执行（Wave 调度 + 契约重试），实测提速 1.4x–2.2x 🚀

Production-ready RAG for codebases using AST parsing, hybrid search, and LLMs with precise file & line citations.

🏆 Companion research environment to the ACM Vibe Science publication · EASE '26 / VibeX · DOI 10.1145/3816483.3816631 · Auditable, resumable AI-assisted research.

Agent skills for building on Vobiz - voice API, SIP trunking, voice XML, and AI voice agents.

Z3-powered reasoning for AI agents. MCP server + CLI for scheduling, logic, config, and safety.

Local-first AI workspace with persistent memory, real tools (files · shell · LSP · MCP · Office), and an opinionated agent team. Your data, your machine.

Open map data for your AI — free, no account, no API key. MCP server for Overture Maps: places, geocoding, routing.

LittleOrange Video MCP 是基于 Model Context Protocol (MCP) 的视频生成服务端，为 AI 应用提供统一、标准化的视频生成 API 调用能力。服务基于官方视频生成 API 文档封装，支持 Sora2、Veo3.1、Vidu Q3、Dreamina Seedance 2.0 等多个主流视频生成模型，内置智能异步轮询、三层工具架构（底层API映射、自动轮询工具、高层Agent工具）、完整素材管理等功能，完美适配各类 AI 客户端与 Agent 场景。

Your coding agent is smart. Your legacy codebase is the problem. Reify gives it the context to understand and safely change large, unfamiliar codebases — 100% locally.

docsforge — Crawler and Adaptive Harvest for Technical Documents like Code Language , API , Technologies. Search -> Identify URL -> Crawl -> Harvest Documentation

An AI-powered web app that helps Indian citizens discover government schemes they qualify for, understand their legal rights in real-life situations, and get step-by-step application guidance — all backed by real-time search from official government portals.

Check detected chart patterns against the rate the market moves that way anyway, so you can see which signals carry any information at all. Lift over a pattern-free baseline with cluster-robust intervals -- never against 50%.

AI-powered IDA Pro REST API with 90+ endpoints. Autonomous RE agent, swarm worker, full SDK coverage.

MCP server for Pi-hole v6 API — supports multiple instances

The personal AI that actually remembers what you teach it. Corrections become lessons + temporal-KG facts retrieved on every future answer; living dossiers distill 70+ monitors of news into durable understanding. Measured, not asserted. Fully local.

From engineering intent to accepted work: native decomposition, HMAC-sealed atomic task contracts, bounded TaskMesh execution, and verifiable acceptance across Codex, Claude Code, Grok Build, Kimi, and Cursor.

Open-source Python framework to build, orchestrate and deploy AI agents: multi-provider, tool calling, RAG memory and desktop distribution.

A file-based governance harness that turns loose research ideas into proposals that hold up. Makes three things checkable: which page a claim rests on, what would refute it, and what the AI just changed. Human stays the sole adjudicator. Bilingual (EN / 繁中).

DSH 多方群聊协作插件（圆桌）：把不同人设的 AI Agent 拉进同一研讨房间，互相讨论、交叉补充、质疑修正，最终一键总结出综合方案；也支持把工作区已有的对话框拉进来参与讨论，消除信息隔离。支持手动 @ / 全体研讨 / 接力链三种发言模式，含轮次上限、全局暂停等防死循环控制。官方 bundle 插件：dsh plugin --profile web add github:vuchisu069-source/dsh-round-table#main

基于 LangGraph 的多智能体深度研究系统，自动搜索、分析、撰写结构化研究报告。

Open-source MCP supervisor for Codex to delegate, monitor, and manage Grok, Cursor, and DeepSeek coding agents.

[ASE 2026] MultiFixer: A Coordinator-Proposer Based Multi-Agent Framework For Fixing Multi-Hunk Bugs

Self-hosted multi-agent workflow orchestration for Feishu/Lark with authenticated dispatch, retries, file intake, and traceable runs.

OMATS: A reproducible benchmark measuring LLM performance in multi-agent environments. Based on the 5 Stages of Model Capability framework.

通用 Agent Skill + CLI，将网页版 AI 作为外部 Agent，用于调研、评审与方案讨论，帮助降低宿主上下文占用和 API Token 消耗；当前仅支持 ChatGPT 自动化。

An adversarial feedback swarm approach to AI-first engineering

dsh-delegate: model-aware subagent delegation for DeepSeek Harness — per-call models, depends_on dependency gating, per-child personas, a durable run roster, audit events, and conversation-flow tool cards. | 给 DeepSeek Harness 的子代理委派加上：按次选模型、依赖门控、角色人设、任务花名册。

Async LLM context relay between DeepSeek Harness sessions and Agent Team Rooms

Capability-first vertical agent framework with policy, approvals, workflows, tracing, evals, and an interactive visual lab.

AI-native desktop command center built on top of DeepSeek Harness

天机(Tianji):多 CLI 编程助手编排框架——六角色机械流水线(总控/分配器/监控器/架构师/实施者/审核者),账本单一真源,机制化不靠自觉

Forge – The AI Execution Layer

Visual MCP gateway composer for Model Context Protocol (MCP): discover and combine MCP servers and tools, manage permissions, validate integrations, and export secure configs for AI agents and LLM clients.

Lightweight Multi-Agent Orchestration Framework for Java

Kuber Map for FIBO — a curated, learner-first map of the Financial Industry Business Ontology that doubles as audit-ready grounding for financial AI agents. 5 use cases, grounded eval (+44.5pt), MIT.

Open framework for building an agentic SEO app for Shopify — audits, JSON-LD automation, AI-search readiness. MIT.

Self-hosted AI assistant with voice interface, persistent memory, and multi-step reasoning. Built on Python with local-first architecture — no cloud dependency, no data leakage.

Official implementation of DoLQ, a multi-agent LLM framework for discovering physically plausible ODEs from observational data.

The first reference monitor for autonomous AI penetration testing agents. Mandatory enforcement, not prompt-based guardrails. Patent pending.

AI agent that recovers failed payments: root-cause diagnosis (rules-first + LLM tail), EV-ranked interventions, deterministic guardrails, payment-link execution — every action on a hash-chained ledger. Pre-registered eval: +10.24pp recovery vs tuned-naive baseline.

Cost-aware multi-agent LLM orchestration with dynamic communication topology, per-round agent selection and a self-halting manager.

🧙‍♂️ Local-first autonomous data analysis agent. Private, self-correcting, and evidence-backed analytics powered by DuckDB, Polars & local LLMs.

Hire AI agents, give them jobs, watch them work. A local-first console where agents own a queue, use real tools, and pass a quality gate before any task counts as done. Zero dependencies, runs fully offline.

a demo plugin for a preset with builtin musescore wasm editor/preview engine for agent composing

Coding on a budget. Free. Cheap.Curated list of open-source tools, free-tier strategies, and local runtimes for cost-efficient AI- assisted coding. Covers LLM model selection, API gateway routing, token compression (up to 90% reduction), MCP context servers, agent quality guardrails, self-hosted inference, and pricing intelligence with grants.

On-device RAG on Android — notes app with local embeddings, vector search, and agentic tool calling. Fully offline, Gemma 4 + LiteRT-LM

Ditch Community Edition is a local macOS workspace for running, following, and returning to Codex sessions across multiple projects

Agentic macOS environment for scientific research — coding agents, papers, figures, LaTeX and project files in one shared context. Rust-first, Tauri 2.

Self-hosted HTML knowledge workspace. Notes are HTML + YAML on disk — import, filter, read, edit, and share. No database, no in-app login, no AI.

✨ Plan your time with calmness, optimize schedules, and manage habits with Google Gemini AI. Open-source smart personal planner & calendar.

QVS-RAG

MCP server with Brazilian public data: IBGE, Central Bank, INMET, Chamber of Deputies and Federal Senate

ctx — a queryable code graph for coding agents, across Rust, Python, TypeScript and Go. Call graphs, blast radius, breaking-change gates, port parity. One binary, no LSP, no index.

Free Web Search for AI agents. Search and fetch the public web with DDGS, SearXNG & Jina — no paid search API required. Reduce Tavily/Firecrawl credits and cut agent research costs.

A Linux-only, ncurses based TUI AI coding agent.

One-pass agentic inbox triage: classify, summarise, extract tasks and draft replies in a single run. Draft-only — no send scope exists, enforced by test. BYOK across 6 LLM providers incl. fully-local Ollama.

Verified DFT workflow MCP toolchain: dual-engine (GPAW/QE), convergence gates that actually gate, runtime results-vs-Materials-Project reconciliation, full evidence chain.

Curated Agentic AI resources: AI agent frameworks, tutorials, courses, projects, MCP tools, research papers, evaluation, security, and production guides.

Drop-in alignment rules for any AI agent (Claude Code, Cursor, GitHub Copilot, OpenClaw, Hermes or a plain LLM system prompt): the eight constitutional clauses of Laukkonen et al. (2025) for four contemplative axioms, quoted verbatim, with an Iterated Prisoner's Dilemma benchmark to measure what a prompt changes.

Official repo for of the ACL 2026 paper "Peering Behind the Shield: Guardrail Identification in Large Language Models"

Architectural experiment: one-sentence to full novel in minutes. Layered state machine + Multi-Head parallel audit. Inspired by diffusion models and Transformer attention. MIT.

跨产品 Agent 任务接力与委派。本机环境适配、额度切换、静默/可见 invoke。Claude / Grok / Kimi 等。

Same-runtime cross-Session discovery and communication for DeepSeek Harness

Autonomous multi-agent due-diligence desk — LangGraph orchestration, agentic hybrid RAG, MCP tools, and a self-improving LLM-Ops loop.

Local-first AI project memory and control runtime for evidence, privacy, model routing, human approvals and cross-harness handoffs.

Your AI coding crew in one Windows terminal - run Claude Code, Codex, opencode & Antigravity as a real team (lead + workers) in a native psmux cockpit. Event-sourced, no WSL, no API keys.

Web dashboard for AgiMate: build agents, set what each one is allowed to touch, and watch the logs

AI-native desktop command center built on top of DeepSeek Harness

Lightweight terminal coding agent - drives a small local model or a frontier model with minimal context usage

Talk Flow AI is a cutting-edge administrative platform designed to manage and analyze AI-driven voice interactions and claims workflows. Built with a focus on speed, aesthetics, and data-driven insights, it provides a seamless experience for administrators to monitor performance.