The OpenAI Agents SDK is a lightweight yet powerful Python framework for building multi-agent workflows. It supports OpenAI APIs and over 100 other LLMs, enabling complex agent collaboration and task automation through core concepts like agents, tools, guardrails, and session management. Key features include agents as tools, sandbox execution, human-in-the-loop mechanisms, conversation history management, and built-in tracing, supporting the development of voice agents.
AI-Search-Hub by minsight-ai-info is an open-source AI Skill designed to aggregate the native search and web extraction capabilities of leading AI platforms like Gemini, Grok, Doubao, Yuanbao, and Kimi. It establishes an efficient, reusable information hub that resolves common data acquisition challenges such as crawler maintenance, platform anti-bot measures, and data cleaning. Users can perform "one query, whole-domain searches" to freely and efficiently gather diverse information, including tech trends, industry sentiment, and data from hard-to-reach sources like WeChat Official Accounts and Douyin. This hub empowers Agents and workflows to leverage major AI platforms' robust understanding and processing power, automating information collection, cleaning, and organization into structured outputs, significantly enhancing data acquisition efficiency and quality.
A layered memory system for AI Agents that combines symbolic short-term memory with hierarchical long-term storage. It optimizes performance by offloading verbose tool logs into compact Mermaid symbols, reducing token usage by up to 61% while improving accuracy. The system organizes experiences into a semantic pyramid (L0-L3), enabling structured persona management instead of flat vector retrieval. Integrated with OpenClaw and Hermes, it provides a traceable and lossless memory architecture.
caveman, developed by JuliusBrussee, is an agent skill/plugin designed to drastically reduce LLM output tokens by adopting a 'caveman-speak' style, while maintaining full technical accuracy. It cuts approximately 75% of output tokens and increases response speed by about 3x. It also features an input compression tool reducing input tokens by around 46%. The product supports various intensity levels (Lite, Full, Ultra, Wenyan mode) and provides specialized commands for terse commit messages and one-line code reviews.
Browser-Use is a powerful AI-driven browser automation tool that enables users to automate web interactions, data extraction, and complex tasks via natural language or Python code. It offers an open-source agent for local deployment and a more robust cloud agent service, featuring stealth browsing, proxy rotation, and advanced integrations to significantly enhance automation efficiency and accuracy.
MemPalace is a local-first AI memory tool designed to store conversation history verbatim and retrieve it through semantic search, without summarizing or paraphrasing. It utilizes a structured index, categorizing information into "wings," "rooms," and "drawers" to enable scoped searches. The retrieval layer is pluggable (ChromaDB by default), ensuring all data remains on the user's machine unless opted out. It achieves an impressive 96.6% R@5 raw recall on the LongMemEval benchmark, notably without requiring LLMs or API calls, showcasing high performance in a local environment.
Everything Claude Code is a performance optimization system for AI agent harnesses, developed by an Anthropic hackathon winner. It offers a complete agent ecosystem including production-ready skills, instincts, memory optimization, continuous learning, security scanning, and research-first development. The system supports various major AI agent harnesses like Claude Code, Codex, Cursor, OpenCode, and Gemini, and has been iterated and validated through over 10 months of intensive daily use. It features a Tkinter-based desktop dashboard and supports multi-language rules and agents.
rtk is a high-performance command-line interface (CLI) proxy designed to significantly reduce LLM token consumption by filtering and compressing command outputs, typically achieving 60-90% savings. Implemented as a single Rust binary, it supports over 100 common CLI commands and employs four core strategies: smart filtering, grouping, truncation, and deduplication. With a transparent auto-rewrite hook for Bash commands, rtk optimizes the context provided to LLMs without altering user workflows, thereby enhancing efficiency and cost-effectiveness when interacting with AI tools.
Onyx is an open-source AI platform designed as the application layer for Large Language Models (LLMs), offering a feature-rich and easily deployable interface. It empowers LLMs with advanced capabilities such as Agentic RAG, deep research, web search, code execution, and file creation. Onyx supports over 50 indexing-based connectors and provides both Lite and Standard deployment modes. It also offers enterprise-grade features like collaboration, SSO, and RBAC, available in Community (MIT licensed) and Enterprise editions, deployable across various environments like Docker and Kubernetes.
LangGraph, from LangChain, Inc., is a powerful low-level orchestration framework engineered for building, managing, and deploying sophisticated, long-running, and stateful multi-agent applications. It represents application logic as cyclic Directed Acyclic Graphs (DAGs), giving developers precise control over LLM execution, including when to pause, loop, or transfer control. LangGraph ensures durable execution, seamless human-in-the-loop capabilities, and comprehensive short-term and long-term memory management. Deep integration with LangSmith provides robust debugging and production-ready deployment, making it the ideal choice for developing complex and reliable AI agents.
Firecrawl is a powerful Web Data API engineered for AI Agents, enabling them to access high-quality, clean web data at scale. It boasts industry-leading reliability, covering 96% of the web, including JavaScript-heavy pages, without proxy headaches. With a blazingly fast P95 latency of 3.4 seconds, Firecrawl transforms web content into LLM-friendly formats like clean Markdown, structured JSON, and screenshots, significantly optimizing AI application inputs. It automates complex tasks such as proxy rotation and rate limiting, and empowers AI Agents with advanced interactive capabilities like clicking, scrolling, and typing. An open-source solution, Firecrawl is essential for building robust, intelligent AI applications.
daily_stock_analysis, developed by ZhuLinsen, is an intelligent stock analysis system powered by large AI models. It automatically analyzes selected A-share, Hong Kong, and US stocks daily, generating a 'Decision Dashboard' with core conclusions, ratings, buy/sell points, risk alerts, and action checklists. The system supports automated delivery of analysis reports via multiple channels including WeChat Work, Feishu, Telegram, Discord, Slack, and email. It integrates multi-dimensional analysis (technical, real-time quotes, chip distribution, news sentiment), supports global markets, and incorporates various built-in market strategies (e.g., moving averages, 'Chan Lun', Elliott Wave). Users can also utilize a web-based workbench for manual analysis, configuration management, historical report viewing, backtesting, and AI Agent-powered strategy inquiries.
AgentScope is a production-ready multi-agent framework designed to simplify the development of LLM-powered applications. It provides high-level abstractions for agent communication, memory management, and tool usage. Rather than using restrictive prompts, AgentScope leverages the inherent reasoning and tool-use capabilities of LLMs. Key features include Real-time Voice, Human-in-the-loop steering, Agentic RL for model tuning, and support for MCP and A2A protocols. It is engineered for scalability, supporting local, serverless, and K8s deployments with integrated OTel monitoring for enterprise-grade observability.
prompts.chat is the world's largest open-source AI prompt library, offering a curated collection of prompt examples for various AI chat models like ChatGPT, Claude, Gemini, and more. It provides an online platform for browsing and contributing prompts, along with interactive learning resources for prompt engineering and a game-based adventure for kids. The platform also supports self-hosting and various integrations including a CLI and Claude Code Plugin.
Jina AI Reader is a tool developed by Jina AI to optimize input for Large Language Models (LLMs). It offers two core functions: the 'Read' feature converts any URL into an LLM-friendly text format, while the 'Search' feature performs web searches and transforms the top 5 search results into an LLM-compatible format. Reader aims to address the challenges of traditional web scraping for LLM input, providing a free, stable, and scalable API service. It supports processing complex content like PDFs, images, and Single Page Applications (SPAs), with fine-grained control via request headers.
Promptfoo is a CLI and library designed for evaluating and red-teaming Large Language Model (LLM) applications. It helps developers build secure and reliable AI apps, moving beyond trial-and-error. It offers automated evaluations, vulnerability scanning, side-by-side model comparison, and CI/CD integration. A key feature is its ability to run evaluations 100% locally, ensuring privacy. It is developer-first, flexible, battle-tested, and provides data-driven insights.
Ragas, an open-source Python library developed by vibrantlabsai, is a comprehensive toolkit for evaluating and optimizing Large Language Model (LLM) applications. It aims to replace subjective and time-consuming assessments with data-driven, efficient evaluation workflows by offering objective metrics (both LLM-based and traditional), intelligent test data generation, and actionable insights. Ragas can automatically create diverse test datasets, seamlessly integrates with popular LLM frameworks like LangChain and major observability tools, and facilitates building feedback loops to continuously improve LLM apps using production data.
nexu is an open-source desktop client that connects OpenClaw AI Agents to IM channels including WeChat, Feishu, Slack, and Discord. It provides a pure GUI experience, allowing users to deploy agents without CLI knowledge. With built-in support for OpenClaw and Feishu skills, it enables agents to be accessible 24/7 via mobile IM apps. It supports various LLMs through OAuth or BYO API keys, maintaining a local-first architecture to ensure data privacy for individuals and teams.
OpenSquilla is a token-efficient, microkernel AI agent designed for CLI, Web UI, and chat channels. At its core, an on-device model router (SquillaRouter) dynamically dispatches tasks to the cheapest compatible LLM (supporting over 20 providers like DeepSeek, Anthropic, OpenAI). Featuring persistent memory, a layered sandbox, web search, and local embeddings, it delivers identical tool dispatch, retries, and decision logging across all interfaces, maximizing efficiency within your budget.
Edict is an OpenClaw-based multi-agent collaboration architecture inspired by the ancient Chinese 'Three Departments and Six Ministries' system. It utilizes 12 specialized agents with strict role-based authorization. Unlike traditional frameworks, Edict features a mandatory institutional review mechanism that can reject unqualified outputs, preventing errors from executing. It includes a real-time Kanban dashboard providing full observability, allowing users to manually intervene in tasks, hot-switch LLM models, and review complete audit logs to eliminate the black-box nature of AI workflows.
TrendRadar is a lightweight, open-source hotspot assistant designed for rapid deployment. It aggregates real-time news via RSS and APIs, using AI for intelligent filtering based on natural language interests, automated translation, and trend summarization. It supports one-click notifications to over 9 platforms, including WeChat, Lark, and Telegram. With deep MCP (Model Context Protocol) integration, it enables AI agents to access and analyze live news data seamlessly, offering a professional HTML reporting interface and cross-platform notification adaptations.
oh-my-pi (omp) is a highly capable, open-source AI coding agent featuring deep IDE integration, built primarily in Rust. Forked from Mario Zechner's Pi, it bridges the gap between LLMs and local environments. It stands out by providing persistent Python and Bun tool-calling kernels, native Language Server Protocol (LSP) integration for exact refactoring, and real DAP debugger attachment (lldb, dlv, debugpy). Additionally, omp introduces 'Hashline' for precise edits, time-traveling stream rules for instant course-correction, and first-class subagent orchestration. Operating natively across macOS, Linux, and Windows, it serves as a comprehensive, out-of-the-box coding assistant.
code-review-graph is an AI-assisted code review tool that builds a structural map of code using Tree-sitter and incrementally tracks changes to provide precise context to AI assistants via the Model Context Protocol (MCP). This approach significantly reduces token consumption by enabling AI models to read only the minimal set of files relevant to a change, rather than the entire codebase. It aims to solve the token waste problem in large repositories and monorepos, improving the efficiency and quality of AI code reviews.
Browser Harness is a thin, self-healing CDP (Chrome DevTools Protocol) tool that connects an LLM directly to a real browser, granting agents complete freedom without rigid middleware frameworks. Its standout feature is self-improvement: during execution, the agent automatically writes missing helper functions on the fly and generates reusable, site-specific domain skills. Operating via a single WebSocket, it supports both local debugging and headless deployment, integrating seamlessly with Browser Use Cloud for stealth capabilities, proxies, and captcha solving.
An educational project by shareAI-lab focused on 'Harness Engineering' for AI agents. It provides a masterclass by reverse-engineering the architecture of Anthropic's Claude Code. The repo teaches how to build the 'vehicle' (harness) for the 'driver' (model), covering essential mechanisms like tool implementation, context compression, subagent spawning, and task dependency graphs. It shifts the development focus from prompt plumbing to building robust environments for autonomous models.
Open Multi-Agent is a lightweight, TypeScript-native multi-agent orchestration framework. By simply providing a goal, its built-in coordinator automatically decomposes the goal into a task DAG, parallelizes independent steps, and synthesizes the final output. With only 3 runtime dependencies, it seamlessly drops into any Node.js backend. Key features include mixing over 12 LLM providers within a single team, a sandboxed filesystem workspace, pluggable shared memory, and native integration with the Model Context Protocol (MCP) for tool extensibility.
DeepSeek-Reasonix by esengine is an open-source, DeepSeek-native AI coding agent built for the terminal and desktop. Its core innovation is a cache-first architecture engineered around prefix-cache stability. By maintaining an append-only loop, it achieves near 99.8% cache hit rates during extensive coding sessions, slashing API token costs by up to 5x. Alongside its robust CLI, Reasonix features a native Tauri desktop GUI and remote QQ channel integration. It fully supports the Model Context Protocol (MCP) for tool expansion, local semantic indexing, web search, and custom Markdown-based skills, making it a highly optimized companion for developers.
Rowboat is an open-source, local-first AI coworker designed to transform daily work content, such as emails and meeting notes, into a persistent knowledge graph. It securely stores all data in Markdown format on the user's local machine, ensuring privacy and data control. By accumulating context and understanding relevance, Rowboat efficiently drafts emails, prepares for meetings, and generates documents and presentations (including PDFs). It also features "live notes" to track specific people, companies, or topics. Rowboat supports various AI models (local and hosted) and external tool integrations via the Model Context Protocol (MCP), enabling advanced workflow automation.
oMLX is a high-performance local LLM inference server optimized specifically for Apple Silicon Macs. Built on the MLX framework, it supports text LLMs, VLMs, OCR, and embedding models. It features a unique Tiered KV Cache system (Hot RAM + Cold SSD) that persists context across requests and server restarts, making it ideal for coding agents like Claude Code. oMLX provides a native macOS menubar app and a web dashboard, supporting continuous batching, automated memory management via LRU eviction, and seamless MCP integration.
lowfat is a lightweight CLI tool designed to reduce AI token costs by filtering out unnecessary and verbose command-line output before it reaches your AI agents. It intercepts common commands like git, docker, and ls, condensing the raw text without losing core signals. Key features include local-first execution with zero telemetry, UNIX-style composability, and customizable compression levels (lite, full, ultra). It seamlessly integrates with AI agent environments such as Claude Code, OpenCode, and Pi.
whichllm is a CLI tool that auto-detects your hardware to recommend the best HuggingFace LLMs that actually fit and run efficiently. Rather than just finding the largest model that fits in VRAM, it intelligently ranks models using real benchmark scores, quantization penalties, and estimated inference speed. It also features GPU simulation, reverse hardware planning, one-command chat execution, and Python snippet generation.
