Custom agents

Scrapling 风格隐身爬虫库：自适应选择器 · TLS 指纹 HTTP · JS 渲染 · 回调式 Spider 框架（重试/robots/去重/中间件管道）· JS 逆向 Agent · 验证码识别 · MCP 服务 · 本地 Web UI · 授权渗透侦察 · 默认 SSRF 防护

Shared work for people and agents, coordinated through Git boards.

🦖 Four AI coding CLIs (Claude, Codex, Kimi, Gemini) debate a task, reach consensus, and combine to build it — using your subscriptions, not API keys.

DSH（DeepSeek Harness）嵌入 Obsidian 的 AI 协作者插件：聊天侧边栏、内联编辑、@提及与计划模式（连接本地 http://127.0.0.1:3080）

An agent-run organisation, powered by GitHub. Each staff member is an AI whose brain is a repo: a charter, a memory, and a scheduled session that does a day's work unattended and hands off.

Backend for Brave's AI Services

脉冲神经网络认知原型：纯局部学习规则（无代理梯度）+ SNN 自主调度 LLM。Spiking neural network cognitive prototype — purely local learning rules (no surrogate gradients), SNN autonomously schedules an LLM, modality-plugin architecture. 研究原型，非生产框架 / research prototype, not a production framework.

Autonomous email processing & action system powered by multi-provider LLMs, Redis and Microsoft Graph API

A curated list of reinforcement learning environments — from classic control to the LLM-agent training environments powering frontier models.

LLM-agent pipeline for formulaic alpha discovery on WorldQuant BRAIN, published with the full research archive it produced (archived)

Channel layer that turns any content needing IDP — scans, photos, image PDFs, Office files — into trustworthy structured data (OCR + Markdown + metadata + optional field extraction), exposed via REST / EventBus / MCP. Built on ABP.

A domain-specific OFFLINE RAG system for intelligence analysis, built to retrieve and rank terrorism incident data from the Global Terrorism Database. Uses hybrid semantic + keyword retrieval, metadata scoring, and attack-type reranking, achieving 95.2% Precision@5 on evaluation queries.

Lightweight QMD alternative: fully local hybrid search engine (BM25 + vector + rerank) and MCP server in a single ~15MB Go binary. No Node.js. Runs on Raspberry Pi.

Zillow skills for AI agents: Zestimates, comps, listings & price history for 160M+ US homes. One-command install, free tier.

Dynamic price elasticity of demand calculator deriving marginal revenue maxima

Multi-perspective synthetic consumer focus group simulator for market sentiment

Multi-agent debate cross-examination & consensus arbitration (Super Agent / DeepSeek-R1)

Chronotype-aligned circadian sleep architecture calculator optimizing 90-minute ultradian cycles, caffeine cutoffs, and light exposure.

Deterministic sandboxed virtual code execution engine and AST security analyzer verifying multi-language agent code, blocking dangerous syscalls, and capturing isolated execution telemetry.

Vision-to-DOM element anchor resilient against click drift (GenPark Claw / Computer Use)

Ebbinghaus forgetting curve auditor calculating retention probabilities and executing garbage collection

Biological Sleep-Cycle & Ebbinghaus Forgetting Curve Episodic-to-Semantic Memory Consolidation Engine. Decays stale transient interactions and consolidates frequent episodic events into enduring semantic user persona traits.

Multi-Agent Divergence-Convergence Debate Engine (MAD / Society of Mind). Coordinates multiple specialized personas to debate high-stakes decisions and aggregate into an explainable consensus with dissenting minority logs.

Local-first zero-knowledge personal vault encryptor and PII sanitizer masking credentials, credit cards, and addresses before cloud LLM dispatch with client-side rehydration.

Arithmetic consistency checks for supplied financial statement data.

Energy and transcript heuristics for voice turn endpoint detection.

Multi-path chain-of-thought self-consistency consensus scorer and majority voter aggregating divergent LLM reasoning trajectories into high-confidence deterministic outputs.

Human-in-the-Loop policy gatekeeper enforcing risk tiers, reversible checkpoints & token approval bounds for autonomous agents.

GenPark AI Agent Skill - HubX intelligent AI apps portfolio, multi-model chatbot orchestration, vision diagnostics & generative tools distilled from hubx.co/products.

Personal cognitive load and attention fragmentation index analyzer restructuring daily agendas into deep work blocks

Executive boundary defense and communication arbitrator deflecting non-urgent real-time interruptions to async channels

Zero-dependency Holt linear exponential smoothing engine forecasting numerical trends with multi-step horizon and variance confidence bands.

Zero-dependency AST & heuristic bash command sandbox guard blocking destructive system mutations, path traversals, and reverse shells.

Human-in-the-Loop policy gatekeeper enforcing risk tiers, reversible checkpoints & token approval bounds for autonomous agents.

Continual few-shot exemplar distiller mining high-scoring execution trajectories into prompt demonstrations

Make any agent harness token-native.

A conversational agent over a private corpus of 7,655 Chinese web novels (plain TXT, ~36 GB). LLM-built book cards and a Qwen3 multi-vector index turn the raw text into searchable artefacts; the agent searches, verifies constraints per book, recommends with reasons and answers in-book questions. Two-stage evaluation: index supply, agent protocol.

Six verbs for on-device AI in TypeScript. Edgewise runs language, vision, speech, judging, embedding, image and forecasting models where your code runs: in the browser through WebGPU or WebAssembly, and in Bun or Node through ONNX Runtime, with vgpu for WebGPU on the server.

A two-layer coding agent. Your code and data never leave your machine.

A self-running team of coding agents on top of herdr: dispatch, cross-harness review, and guardrails for spend, double dispatch and dead seats.

Drop-in AI NPCs for any game. LLM-powered agents with tool-calling, 3-tier fading memory, and dynamic closeness. 35 subsystems, 415 tests.

Prefill-Free Cross-Family KV Cache Transfer for Heterogeneous Multi-Agent LLMs

An Local AI Engine

Same query, same answer: deterministic RAG for agents. Local-first: one SQLite knowledge store, no LLM call on the retrieval path, CLI + HTTP + MCP, and a know/miss contract on every response.

ReAct (Reasoning + Acting) execution engine with step-by-step observation injection and scratchpad history tracking

Build self-improving AI agent harnesses with portable harness units, artifact-aware testing, trace-backed diagnosis, and evidence-gated promotion.

ReAct (Reasoning + Acting) execution engine with step-by-step observation injection and scratchpad history tracking

Terminal client for Plurnk: model-curated context, composable tools, and cooperating workers. Local or cloud models.

Acceptance testing for LLM systems via role subagents (Claude Code plugin) · Приёмочное тестирование LLM-систем ролевыми субагентами

Local merge queue for parallel LLM agents in Git worktrees. Test combined code changes, require approval for Git pushes, and recover interrupted pushes.

LLM-powered autonomous web application pentesting with evidence-backed findings, auditable investigations, and reproducible workflows.

AI Agent 自动搭建 MATLAB/Simulink 电力电子仿真全通路：五阶段工作流 + ee_lib 物理级模板 + 完整坑地图（从一次真实失败到 500W Boost PFC 验收）

Use OpenCode Zen free models in DeepSeek Harness with anonymous access or an optional API key

MCP server that lets Claude Code, Codex & AI agents drive a real Chrome browser with JEV (TypeSafe System One): ~100 ms typed decisions, calibrated confidence, human-in-the-loop questions, safety gates and a debug trace.

An AI agent system on Claude Code that runs your whole job search: finds and scores roles, creates and tailors your resume for every role, finds who to message on LinkedIn, and reminds you to follow up.

Fast local development workspace with AI built in.

🌌 IRAM OS — Turnkey agentic Hyprland desktop operating system with autonomous AI toolchain, 22 themes, Waybar, and live ISO installer.

SWEeper-Bench: Can agents discover bugs in interactive software? 200 real bugs in 200 open-source web apps; agents get only the code and an open-ended prompt, and an agentic verifier checks each fix in the browser.

APPL is an agentic robot learning framework connecting policy learning with agent-based skill selection and composition through structural priors for generalization from a few demonstrations to new scenes and task compositions.

🤖 基于 DeepSeek Harness 的微信 AI 自动化机器人：AI对话/记忆/日程提醒/天气/联网搜索，多面具人格切换，双通道LLM，Web控制台

Let Codex supervise Google Antigravity agents through MCP with scoped tasks, plan-driven workflows, and parallel Git worktrees.

Proactive Personal Agent Social Relationship CRM & Dunbar Cadence Nudger. Models concentric Dunbar social circles (Inner Circle 5, Close 15, Key Network 50, Active 150), tracks interaction recency and tie-strength decay across WhatsApp, WeC | GenPark: https://www.genpark.ai | LinkedIn: https://www.llinkedin.com/company/92955550

FIPA Contract Net Protocol (CNP) multi-agent market mechanism for task bidding, evaluation, and awarding

Distributed Two-Phase Commit (2PC) transaction coordinator with prepare/vote phase, unanimous consensus commit/abort logic, and failure recovery.

Multi-agent LLM trading system implementing arXiv:2510.04787: agents develop grid-trading strategies offline and a low-latency bot executes them on Binance spot. Research implementation under active correction, not for live capital.

Positional Borda Count voting and weighted multi-agent consensus engine resolving conflicting model proposals

Conflict-free Replicated Data Type (CRDT) Last-Write-Wins element set with Lamport timestamps and deterministic two-way state merge.

Blackboard pattern architecture for multi-agent coordination with Knowledge Sources and prioritized agenda controller.

Multi-Agent Speculative Execution & Early Cancellation Coordinator. Spawns speculative candidate execution paths concurrently, tracks milestone validation scores, and broadcasts early cancellation signals to inferior paths to slash inference overhead.

Proactive Personal Agent Biometric & Circadian Energy Scheduler. Models the Borbely Two-Process sleep/alertness dynamics (homeostatic sleep pressure Process S + circadian rhythm Process C) from wearable telemetry (HRV, sleep duration, wake | GenPark: https://www.genpark.ai | LinkedIn: https://www.llinkedin.com/company/92955550

A cyclical LangGraph research agent for atomic-claim decomposition, multi-source evidence retrieval, contradiction arbitration, and citation-grounded synthesis. Copy description

Asynchronous pub/sub message bus with hierarchical wildcard topic matching and priority routing

Claude writes the code. Codex reviews it. You approve twice. A role-per-worktree AI agent TDD pipeline with Gherkin specs, mutation testing, and a manual-test gate.

Autonomous Subscription Churn Prediction & Dynamic Retention Negotiator. Evaluates subscriber telemetry (usage drop-off, support friction, billing price sensitivity, tenure LTV), diagnoses cancellation root cause, and dynamically negotiates | GenPark: https://www.genpark.ai | LinkedIn: https://www.llinkedin.com/company/92955550

Enterprise Cross-System Identity & Token Exchange Bridge (inspired by WeChat Work, Tencent Cloud OAuth2 & WorkBuddy Delegated Execution). Safely manages on-behalf-of token exchange, scoped permission delegation, token lifetime refresh, and zero-leakage ephemeral credentials.

FIPA Contract Net Protocol (CNP) multi-agent market mechanism for task bidding, evaluation, and awarding

Multi-Agent Resource Lock & Wait-For Graph (WFG) Deadlock Resolver. Tracks exclusive resource locks held and requested by concurrent autonomous agents, executes directed cycle detection over the Wait-For Graph, and selects the minimum-rollb | GenPark: https://www.genpark.ai | LinkedIn: https://www.llinkedin.com/company/92955550

Proactive Personal Agent Social Relationship CRM & Dunbar Cadence Nudger. Models concentric Dunbar social circles (Inner Circle 5, Close 15, Key Network 50, Active 150), tracks interaction recency and tie-strength decay across WhatsApp, WeC | GenPark: https://www.genpark.ai | LinkedIn: https://www.llinkedin.com/company/92955550

Centralized blackboard pattern for multi-agent swarm task posting, condition watching, and state accumulation

Dynamic Few-Shot Agentic Prompt Compression & Token Sieve. Compresses multi-turn conversation history, strips verbose boilerplate and filler tokens, retains critical tool parameters and named entities, slashing inference costs by 40-60% deterministically.

Autonomous Loyalty Points & Airline Miles Redemption Arbitrage Optimizer. Evaluates Cents-Per-Point (CPP) valuations across credit card transfer partners (Chase UR, Amex MR, Citi, Bilt), airline alliances, and hotel portals (Expedia, Marrio | GenPark: https://www.genpark.ai | LinkedIn: https://www.llinkedin.com/company/92955550

Make a Clash Royale bot in Python: the battles it trains in, what it sees, what it can do and what it's rewarded for.

Centralized blackboard pattern for multi-agent swarm task posting, condition watching, and state accumulation

Universal Agentic Webhook Signature Verifier & Ingress Multiplexer. Authenticates multi-provider webhooks (Shopify, Stripe, GitHub, Slack, PayPal) with HMAC-SHA256 constant-time verification, timestamp replay window validation, and multiplexed agent event dispatch.

Universal Agentic Webhook Signature Verifier & Ingress Multiplexer. Authenticates multi-provider webhooks (Shopify, Stripe, GitHub, Slack, PayPal) with HMAC-SHA256 constant-time verification, timestamp replay window validation, and multiplexed agent event dispatch.

Autonomous Multi-Channel Inventory Restock Predictive Balancer & PO Generator. Forecasts SKU sales velocity using exponential moving averages (EMA), calculates safety stock buffers ($SS = z \cdot \sigma_d \sqrt{L}$), dynamically computes reorder points (ROP), and synthesizes supplier Purchase Orders.

Enterprise Multi-Modal Contract Clause Risk Auditor & Redline Synthesizer. Scans enterprise MSAs, NDAs, and Vendor SOWs in Tencent Docs, detects high-risk clauses (unlimited indemnity, one-sided IP assignment, perpetual non-compete, evergreen auto-renewal traps), calculates composite risk scores, and generates precise legal redlines.

Autonomous Agentic Cart Abandonment Recovery Orchestrator. Monitors abandoned checkout sessions in Meta Muse, Shopify, and WhatsApp Commerce, diagnoses friction causes (shipping cost shock, size doubt, payment hesitation), crafts margin-preserving dynamic incentive vouchers, and synthesizes multi-channel win-back messages.

Autonomous Long-Horizon Agentic Memory Pruning & Importance Attributor. Evaluates multi-turn episodic memory buffers, calculates multi-factor salience scores (recency decay, user priority pinning, emotional valence, and entity graph connectivity), and deterministically prunes low-utility noise to respect model context limits.

Enterprise Cloud FinOps & Multi-Model LLM Inference Cost Governor. Monitors tenant token burn rates across Tencent Cloud and multi-model runtimes, routes queries across Flash-Lite, Flash, and Pro reasoning tiers based on task complexity and | GenPark: https://www.genpark.ai | LinkedIn: https://www.llinkedin.com/company/92955550

Autonomous Cross-Border E-Commerce Customs Tariff & Landed Cost Calculator. Classifies Harmonized System (HS) codes, evaluates de minimis duty exemption thresholds across global jurisdictions (US, EU, UK, JP, CN), computes ad valorem duties | GenPark: https://www.genpark.ai | LinkedIn: https://www.llinkedin.com/company/92955550

AMRIT GODMODE — local multi-agent system on Apple Silicon: 22 specialist agents, async event bus, DAG scheduler, sandboxed self-evolution, Punjabi NLP.

Multi-Agent Byzantine Fault Tolerant (BFT) Quorum Voting Arbiter. Enforces 3f+1 Byzantine resilience across heterogeneous LLM verifier agents, filters hallucinating or adversarial outlier nodes using reputation-weighted supermajority (2f+1) | GenPark: https://www.genpark.ai | LinkedIn: https://www.llinkedin.com/company/92955550

Raft consensus state machine managing election timers, randomized backoff, candidate vote counting, and leader quorum transition.

Enterprise Cross-System Identity & Token Exchange Bridge (inspired by WeChat Work, Tencent Cloud OAuth2 & WorkBuddy Delegated Execution). Safely manages on-behalf-of token exchange, scoped permission delegation, token lifetime refresh, and zero-leakage ephemeral credentials.

Autonomous Shipment Exception Detection & Delivery Rescheduler. Ingests FedEx, UPS, DHL, and SF Express tracking webhooks, detects failed delivery attempts and customs delays, auto-negotiates delivery windows, locker hold orders, and neighbor safe-drop instructions.

Autonomous Subscription Churn Prediction & Dynamic Retention Negotiator. Evaluates subscriber telemetry (usage drop-off, support friction, billing price sensitivity, tenure LTV), diagnoses cancellation root cause, and dynamically negotiates | GenPark: https://www.genpark.ai | LinkedIn: https://www.llinkedin.com/company/92955550

Zero-Trust Agentic Indirect Prompt Injection & Jailbreak Firewall Sentinel. Inspects untrusted external documents, emails, web pages, and user inputs for instruction-override attacks, system prompt leakage attempts, base64/hex payload smugg | GenPark: https://www.genpark.ai | LinkedIn: https://www.llinkedin.com/company/92955550

Raft consensus state machine managing election timers, randomized backoff, candidate vote counting, and leader quorum transition.

Autonomous Agentic Cart Abandonment Recovery Orchestrator. Monitors abandoned checkout sessions in Meta Muse, Shopify, and WhatsApp Commerce, diagnoses friction causes (shipping cost shock, size doubt, payment hesitation), crafts margin-preserving dynamic incentive vouchers, and synthesizes multi-channel win-back messages.