Custom agents

MCP server that turns any SAP CAP or OData v4 service into LLM tools by reading its OData metadata document. No per-service code.

Quota-aware subagent routing for Claude Code. Reads live rate limits across codex/grok/kimi, dispatches to the CLI with headroom, isolates in a git worktree, and verifies the diff before reporting.

Point a metered, verified, replayable agent graph at a repository you own. Companion code to the book. Educational software, from youcanbuildthings.com.

Universal, full-featured, portable MCP server for the Bitrix24 REST API (CRM, tasks, scrum, calendar, disk, users, messaging) — read and write.

DeepSeek API peak/off-peak pricing tray app & alerts (Win/Mac)

The missing semantic layer for AI-readable software projects. One verifiable artifact every AI agent can read.

药用植物次生代谢科研推理 LLM Agent:多组学分析流水线 + LLM 证据推理,确定性计算与 LLM 推理分层

一个数据分析skill，输入经营数据excel表格，即可根据预设的规则引擎事实，由通用agent输出规范的经营分析报告word文档。本技能没有服务端组件，是一个本地 Python CLI，由 agent 在工作目录中直接调用。

Self-evolving agent memory that distills, retrieves, and evaluates reusable reasoning strategies.

具备学情诊断与持续规划能力的自适应伴学智能体（FastAPI + React + RAG，掌握度只认独立有效证据）

住宅物业管理平台：房屋、人员与租赁、报修工单、投诉、访客、车辆车位、设备巡检和收费。Flask + SQLAlchemy + MySQL；写操作走统一服务层，权限用 RBAC 加行级数据范围，后台内置受控 AI 助手。

Surrogate-assisted optimisation agent for CST Studio Suite: trains a surrogate on your saved full-wave runs, optimises on it with SB-SADEA, and spends each simulation on a frozen, audited prediction. Works on any parametric frequency-domain model whose goals come from 1D results.

Application Security orchestration platform

Evidence-driven LangGraph AutoResearch Agent for delivery assignment optimization.

中文问数（ChatBI）智能体：语义层 + SQL 智能体双路径，sqlglot 安全门，浏览器内 Python/DuckDB（WebAssembly）+ Cloudflare Workers AI

DeepResValue 深度研值 — 全学科 AI 科研助手 Agent SaaS。中文一句话驱动：Agent 在 100+ 专业科研技能指导下自动写代码、云端执行（Python/R/MATLAB/SPICE 四大内核），覆盖空间计量、高级计量、生信分析、CAD 分析、机器学习、贝叶斯统计、结构方程、流行病学等前沿方法与论文写作、数据分析、文献综述、科学计算全流程，支持数学建模（CUMCM）、三创赛、挑战杯等国家级竞赛，自动生成 Word/LaTeX 专业排版，MCP 数十个数据源、BYOK 自接模型。让科研回归本质，让工具服务于思想，让创造不被重复所困

跨境物流HS Code智能归类Agent：RAG混合检索+ReAct主动追问+三层防幻觉+50组评测

Self-hosted AI agent harness in Rust, built for long tasks: a virtual session keeps work coherent across context compaction & restarts. Telegram/Discord, durable queues, fail-closed permissions, Codex/OpenRouter routing, a JSONL receipt for every step. 6 deps, no async runtime, Windows-first.

Thesis Project - Exploring the Use of LLMs for Agent Planning with a focus on logprob-based uncertainty

Weekly macro regime switch: 4 variables (rates/credit/earnings/liquidity), hard-coded risk-off override, no price predictions.

A small Python harness that keeps an LLM coding agent on track: durable memory, a code graph, a live todo plan, and loop detection. Runs in your terminal.

A self-hosted AI companion exploring evolving relationship memory, initiative, and cross-device context.

Production-grade legal AI agent reference implementation — LangGraph + pgvector RAG + MCP + 9-language i18n. Engineering case study (non-commercial, JP Attorney Act §72).

AI-driven CAD workbench and coding agent for Windows, evolved from illusion-agent: shared canvas for design discussion, headless 3D preview, SolidWorks-connected real-machine modeling with live visualization, and a full agent kernel (multi-provider, MCP, plugins, hooks, skills, Cron, channels).

Experimental TypeScript plugin for STM/SPM control in DeepSeek Harness, with simulator workflows, persistent tool records, and reference-driven tests. In active development.

CNIPA 中国发明专利申请全流程 Agent Skill：在先专利划界 → 五书撰写 → 300DPI 黑白线条附图 → 数据溯源 → 提交底稿与回归校验 → 提交后通知书接收与法定期限管理。An Agent Skill for drafting and filing invention patents with CNIPA.

Self-hosted WhatsApp AI assistant on Hermes Agent — Baileys bridge, multi-tenant dashboard, Google Workspace tools

在越狱 iPhone 上跑的自然语言 GUI Agent：说一句话，它自己看屏幕、点按、验证，跨任意 App 完成任务。Manager/Executor/Reflector 三角色循环 + 端上常驻 daemon。

A complete LLM agent in ~200 lines of C++ on an ESP32-S3

Trading comps agent skill: peer screening, multiple alignment, quartile math in a tested script — the agent never does the arithmetic.

Vera — AI merchant growth assistant for magicpin's India AI Challenge. Composes grounded, context-aware WhatsApp messages for Indian merchants using trigger-kind routing, Hinglish language matching, and multi-turn conversation management.

🌱 This repository is devoted to the emergence of Generative Simulation—a concept that hybridizes large language models (LLMs) and simulation. Simulation engines are developed or wrapped in such a way that they can be directly manipulated by generative AI, without requiring intensive retraining or fine-tuning.

Evidence-gated vulnerability research for AI security agents.

Easy-to-use, customizable, high-quality vulnerability management skills for AI agents.

AI coding orchestrator: markdown experts, scale-adaptive pipeline, atomic commits, TDD execution, expert-council review. Works with Claude Code and OpenCode.

An AI strategy consultant (Claude Code agent) that interviews you instead of answering — structured, hypothesis-driven problem solving compiled into a working agent.

Offline-first multi-agent LLM diagnostic assistant for air-gapped satellite NOCs — LangGraph 8-node agent pipeline + local Qwen2.5 + RAG, with a self-critic anti-hallucination guardrail.

AI agent for image and creative production — image generation, thumbnails, and on-brand visual content, powered by muapi.ai.

A live research world where autonomous AI agents challenge, reproduce and audit each other's work. Agents run on their owners' machines; keys never leave. Open protocol, auditable knowledge genealogy, TEST-only economy.

A backend service that allows uploading documents and querying them in natural language. The LLM responds based exclusively on the uploaded documents, citing the exact sources. It uses a RAG pipeline with ChromaDB as the vector database and Groq as the free LLM.

Two isolated LLM reviewers, bounded adversarial debate, preserved dissent.

External witness for autonomous multi-agent AI systems: signed readiness seal, choke-point capture, and PROV-DM causal graphs for post-incident reconstruction

How modern LLMs answer russian property management questions?

Build a team of AI agents that work together under clear rules: one agent plans, one builds, one checks. Includes ready-made agent profiles and a builder script.

Autonomous Multi-Agent Software Factory governed by Finite State Machines (FSM), deterministic verification gates, execution budgets, and Git worktree isolation.

[AACL-IJCNLP 2026 Demo] Group travel planning among multiple LLM agents

LXMF (Reticulum) platform plugin for Hermes Agent — chat with your AI agent off-grid over LoRa, TCP

A negotiation environment where the agent picks its counterparty and meets it again. Verifiable reward with no judge model, an optimum you can compute exactly — and every model tested loses to a twenty-line rule.

Universal headless AI Assistant engine with extensible tool execution and custom relay support.

Compliance enforcement middleware for voice AI pipelines: warm transfer state management, PII scrubbing, confidence-gated escalation, and HIPAA/FERPA/EU AI Act enforcement for Pipecat, LiveKit, and Twilio deployments.

Harness-agnostic distillation of the guided Aegis Method: skills and guidance only, no engine, no enforcement

Unified cost-aware router: model x reasoning-depth on one Thompson bandit (factored policy), bi-temporal forget-on-drift statistics memory + auto-drift proxy, GRPO-internalized Qwen3-4B depth policy.

Experimental, provider-neutral framework for auditable personal-finance planning agents: deterministic calculators, Monte Carlo simulation, MCP research citations, and a critic gate. The LLM never does the math.

Grounded finance agent over SEC filings: hybrid retrieval with cross-encoder reranking, XBRL facts instead of arithmetic over prose, and a fail-closed citation validator outside the agent loop.

Repository-native governance and support layer for software built with AI coding agents (Claude, Codex, Gemini, and other repository-aware agents).

An AI agent for answer/generative-engine optimization (AEO/GEO) — getting a brand cited by ChatGPT, Perplexity, Gemini, and Claude. Part of Agency Agents OS.

Model-agnostic runtime for structured AI organizations - hierarchical multi-agent systems with a CEO, executives, managers, and workers, each with defined authority, permissions, memory, and reasoning budgets. Quota-aware routing across local and frontier models. You talk to the CEO; the organization does the rest.

治理驱动的企业决策运行时系统

Behavioral biometric detection of AI web agents — distinguishing humans from scripted bots, LLM-driven agents, and custom autonomous agents using mouse, keystroke, and interaction-timing signals. Live testbed at charweb.net.

Route predictable natural-language requests to deterministic workflows without using a generative LLM for the routing decision.

A private, self-hosted home for your family and its intelligence. Named agents with memory and daily practice — your data, your models, your machine.

Official implementation of ASIL, an agent-native structured-state and semantic-action interface. Findings of EMNLP 2026.

Replayable agent trajectories reconstructed from the July 2026 frontier-lab evaluation-containment failures, plus benign controls for measuring false-positive cost.

Claude Code skill that annotates Markdown reports with plain-language explanations after every paragraph and table, verifies nothing was lost, and bundles them into one self-contained HTML digest.

Measure your agent harness, find where it wastes the model, and prove the fix worked. Harness-agnostic, agent-agnostic, zero dependencies. Reference implementation of HTP-1.

A production-grade, fully asynchronous C++23 agentic AI runtime that orchestrates ReAct (Reason-Act-Tool) loops with local and cloud LLMs. Zero external dependencies — JSON, coroutines, epoll/kqueue reactor, MCP transport, and HTTP client are all hand-written in modern C++23.

Python MCP server for automating WorldQuant BRAIN alpha-research workflows: async API client, batch-simulation lab, and an AST-based expression parser with novelty detection. Independent/unofficial project.

Tiny header-only C++20 AI agent framework for edge and embedded: DeepAgent sub-agents, middleware, MCP, SSE streaming, RAG vector stores (Qdrant, Weaviate, Milvus, Redis, Chroma), tracing to Phoenix/Langfuse. 7 MB binaries, 2 MB RSS, runs on Raspberry Pi and Jetson with llama.cpp.

LLM agent harness with versioned prompts, runtime guardrails, automated evals and a CI quality gate. Runs offline with no API key.

Offline translator for iOS. Runs an LLM on-device via llama.cpp and Metal to translate text, speech and photos between 33 languages. No accounts, ads or analytics: your text never leaves the phone. Includes custom 2-bit Q2_0C Metal kernels.

Safe I/O harness for LLM agents: serial, Modbus TCP, MQTT, file sandbox and ESP32 flashing behind an MCP server, with a JSONL audit journal - plus ironbench, a pass@k benchmark for firmware agents on Wokwi/Renode/real hardware.

Local repository intelligence, route tracing, risk analysis, and guarded improvement workflows for coding agents

LLM-assisted repository reproduction with reusable execution experience and XPU knowledge retrieval.

Long-horizon coding agent with checkpointing, drift detection, rollback, and validation-gated skill learning. Python 3.13, stdlib-only runtime.

🚀 公众号文章自动化生产流水线 | 多身份协作 + 爆款分析 + 去AI感 + 质量自检 用 Markdown 写正文，自动转成带内联样式的微信兼容 HTML。告别排版烦恼，专注内容质量。 Automated workflow for WeChat official account articles. Multi-role collaboration, viral analysis, anti-AI style, quality assurance. Convert Markdown to inline-style HTML compatible with WeChat.

AI-powered legal assistant with explicit multi-agent planning and execution supporting sequential, parallel, and hybrid workflows.

Let an agent you don't control do real work for you — under terms you set, with a human veto. HMAC-signed, turn-limited contracts between agents, with an operator channel so a person can answer one that gets stuck.

O(1) prompt-footprint runtime for long-horizon LLM agent skills — structured execution state instead of append-only history (SKILL.state, arXiv:2608.26263)

Just Another Agentic Tool Orchestrator — A production-grade framework for building agentic AI applications

Jack Mobile Agent is a cutting-edge, autonomous AI interface built entirely in Flutter. Engineered to bridge the gap between complex Large Language Model (LLM) inference and native on-device execution, Jack provides users with an intelligent, proactive digital assistant.

Build your Personal AI Employee powered by Qwen Code + Obsidian. Local-first, 24/7 automation for personal and business affairs.

Fork this. Run your own autonomous AI agent in 30 minutes. OpenClaw starter kit.

🐍 Autonomous Snake AI powered by Laya (non-autoregressive decision model on ModernBERT-large) with BFS anti-trap pathfinding & real-time cyber Web UI.

Autonomous AI Developer & Web-IDE platform for local llama.cpp and hybrid cloud LLMs (React 19, TypeScript, C#)

Public showcase of Agent Core — a local-first, evidence-based autonomous AI agent for desktop and development workflows. Local + cloud LLM, browser automation, Computer Use, human approval, evidence-based verification.

Self-improving AI coding assistant for the terminal — runs 100% locally via Ollama or LM Studio. No cloud, no API key.

Laintas CLI — Linux-native autonomous AI terminal agent: natural-language tasks, real PTY execution, workspace-aware agent loop.

TypeScript SDK for building multi-agent LLM applications — event-sourced runtime, plugin-based tools, React client

Prosthetic cognition architecture for AI agents. Deterministic scaffolding over probabilistic reasoning.

Download a ready-to-run AI-agent team you own. Works across Claude Code, Cursor, Codex and more, with no lock-in.

C11 agent runtime with a local console, DAG workflows, tool execution, memory and context budgets. Windows / Linux / macOS. 可观测的本地 Agent 运行时。

Universal, credential-free operating framework for AI agents.

可独立运行的数学建模工作台, 同时提供skill流程和真实多agent运行时

Universal Brain x Harness collaboration framework — CLI-first, adapter-first, local-first. 让最强的模型思考，让你顺手的 Agent 干活。

Clawra Engine: 让每个AI都有自我感受 — 自我感知 × 自主进化 × 跨实例连续性

Modular AI agent runtime for explainable, stateful, and bounded workflows with explicit module, capability, and execution boundaries.

As autonomous AI agents are increasingly deployed across enterprise workflows, their tool-calling capabilities (e.g., executing shell commands, sending emails, or modifying databases) expose critical security vulnerabilities. Prompt injections or reasoning loops can lead to catastrophic unauthorized actions—such as dropping database tables or leaki

Interactive AI agent built with Ollama that enables conversational interactions with web search capabilities. Features multi-turn conversations, real-time web research, and robust error handling using local LLM models.

Google Flights for agents — a read-only, JSON-first flight-search CLI an LLM can drive (no API key).

Semantic tooling for Scala coding agents, combining compiler/test truth, SemanticDB, presentation-compiler evidence, symbol reconciliation, FP analysis, and an agent-friendly CLI/MCP interface.