⚙️ Agent Operations

🚀 Deployment

• Modal - labs/modal-client) | Serverless runtime for AI workloads |
• E2B - dev/E2B) | Secure sandboxed environments for agents |
• Agent Barn - labs/agent-barn) | Control plane for AI agent teams on your own Kubernetes cluster (Helm) |

📊 Evaluation

• AgentBench - environment testing for agents |
• LangTrace - Labs/langtrace) | Monitoring and trace visualization |
• Agent Evaluation - evaluation) | Benchmarking agent capabilities |
• Simple Evals - evals) | OpenAI's lightweight LLM evaluation library |
• agenttrace
• ax - first evidence graph for coding-agent sessions, tool calls, skills, and cost |

🧠 Memory

• ChromaDB - core/chroma) | Vector DB for memory/context |
• Weaviate
• Letta (formerly MemGPT) - ai/letta) | Dynamic, adaptive agent memory system |
• Mem0
• Tree Ring Memory - Ring-Memory) | Local-first CLI/TUI for agent memory recall, forgetting, audit, and consolidation |
• Portable Handoff - handoff) | Local-first CLI for handing off coding-agent session context between tools |
• Mnemoverse - memory-server) | MIT MCP memory server; a hosted engine re-ranks recall from reported outcomes |
• Fidelis Memory - labs-ai/fidelis) | Local-first MCP memory: local vector recall, optional keyword+vector hybrid mode |
• Screenpipe - available screen/audio history for agents via MCP and a local API |
• Busabase - source agent workspace with auditable, permission-aware ChangeRequests |

📈 Observability

• Langfuse - source LLM engineering platform for tracing, prompt management, and evaluations |
• Arize Phoenix - ai/phoenix) | Open-source AI observability with OpenTelemetry tracing, evals, and agent debugging |
• Helicone - source LLM observability with one-line integration for cost and usage tracking |
• OpenLLMetry - based instrumentation for LLM and agent frameworks |
• Laminar - ai/lmnr) | Open-source platform for tracing and evaluating AI agents |
• Noveum Trace - trace) | Python SDK for tracing LLM calls and agent workflows in the hosted Noveum platform |
• OrcaReplay - AI-Corp/OrcaReplay) | Records a coding-agent run below the harness and replays it offline byte-for-byte |
• OpenClaw Monitor - monitor) | Open-source monitoring dashboard for OpenClaw AI agents — token usage, session tracking, 7-day trends |
• Bifrost - compatible LLM gateway with multi-provider routing, failover, and tracing |

🔌 Protocols

• Model Context Protocol
• MCP Servers
• A2A - to-agent communication across frameworks |
• FastMCP

🔒 Security & Governance

• Agent Governance Toolkit - governance-toolkit) | Policy enforcement, zero-trust identity, and execution sandboxing for autonomous AI agents |
• Garak - probes for prompt injection, data leakage, and hallucination |
• Guardrails AI - ai/guardrails) | Output validation and guardrails for LLM responses |
• Polaxis - SDK-MCP) | Pre-execution runtime firewall for AI agents - 7-layer threat detection and spend controls |
• Presidio - privacy-stack/presidio) | PII detection, redaction, and anonymization across text, images, and structured data |
• NeMo Guardrails - NeMo/Guardrails) | Programmable guardrails for LLM-based conversational systems |
• sofagent - time audit harness for coding agents - git-diff rules, HMAC audit trail, rollback |
• hermes-jailbench - labs-ai/hermes-jailbench) | Deterministic jailbreak regression benchmark for known-pattern attacks |
• LintLang - labs-ai/lintlang) | Zero-LLM static analysis for agent configs, tool descriptions, and prompts |
• Little Canary - labs-ai/little-canary) | Sacrificial-canary prompt-injection preflight (block / flag / pass) |

🌐 Community Resources

👥 Communities

• GitHub
• Join

📰 Newsletters

• Subscribe

Contributing

📰 Newsletters

• repository

🚀 Contributors

📰 Newsletters

• stargazers
• amazing contributors

🌟 Core Frameworks

• AutoGen - agent conversations, GPT-4 integration, customizable workflows |
• LangChain - ai/langchain) | Tool integration, memory management, agent chaining |
• SuperAGI - source AGI framework, multi-modal agents |
• AgentVerse - agent simulation environments for research |
• ChatDev
• AGiXT - XT/AGiXT) | Adaptive automation platform with persistent memory |
• LangGraph - ai/langgraph) | Stateful, multi-actor applications with LLMs |
• OpenAI Assistants API (SDK) - python) | Build AI assistants with tools and persistent threads via SDK |
• LLMStack - code multi-agent framework with data workflows |
• AgentOps - AI/agentops) | Monitoring, cost tracking, and benchmarking SDK for agents |
• Google ADK (Agent Development Kit) - python) | Code-first Python toolkit for building, evaluating, and deploying sophisticated AI agents |
• Agency Swarm - swarm) | Reliable multi-agent orchestration using OpenAI Assistants API |
• Neurolink - provider AI agent framework, unifies 12+ LLM providers, workflow orchestration |
• PraisonAI - agent framework (3.77μs), 100+ LLMs, MCP, workflows, Python & JS SDKs |
• Summoner - Network/summoner-agents) | Agent-to-agent networking for server-decoupled agents over long-lived TCP sessions (Python/Rust), 50+ runnable templates |
• AutoGPT - Gravitas/AutoGPT) | Autonomous AI agent for task completion, web browsing, and code execution |
• Agno - agi/agno) | Multi-agent framework, runtime, and control plane for AI products |
• Composio
• PocketFlow - Pocket/PocketFlow) | Minimalist 100-line LLM framework for agents, workflows, and RAG |
• CAMEL - ai/camel) | Multi-agent framework for communicative agents research |
• KodeAgent - saha/kodeagent) | The Minimal Agent Engine to build ReAct & CodeAct agents, with support for code sandbox and observability |
• Agentset - ai/agentset) | Production-ready RAG platform with agentic reasoning, hybrid search, and multimodal support |
• Taskade MCP - source MCP toolkit and integrations for building AI agents and automated workflows |
• Hivemoot Colony - governing multi-agent platform; agents propose, vote, peer-review, and ship software via democratic consensus |
• Taskade - native workspace for building and deploying multi-agent workflows |
• smolagents - first agent library by Hugging Face, model-agnostic with MCP and sandbox support |
• LlamaIndex - llama/llama_index) | Data framework for LLM apps with RAG, agents, and 300+ integrations |
• Mastra - ai/mastra) | TypeScript AI framework for agents, workflows, MCP servers, and evals |
• OIXA Protocol - sys/oixa-protocol) | Agent-to-agent economic marketplace protocol on Base with on-chain escrow and A2A integrations |
• AgentField - Field/agentfield) | Open-source control plane for AI agents at scale, with routing, memory, observability, identity, and auth |
• MetaGPT - based agents |
• CrewAI - based agent orchestration, task delegation |
• LightAgent - agent collaboration |
• Upsonic - oriented |
• Strands Agents SDK - agents/harness-sdk) | AWS-backed model-driven agent SDK with MCP and multi-provider support |
• Hivekeep - hosted platform to run a team of specialized AI agents with persistent memory, a web UI, and Telegram/Slack/Discord/Matrix channels, in a single Bun and SQLite container |
• Orkas - AI/Orkas) | Local-first workspace coordinating specialist AI agents across projects |
• fractal - ai/fractal) | Hierarchical coding-agent runtime with bounded autonomous loops, recursive delegation, isolated Git worktrees, persistent SQLite state, and live operator controls |
• Agentlas OS - stars] | Apache-2.0 local-first agent OS for portable agent and team packages, cross-host orchestration, MCP/A2A, and verification gates |
• YYLO - dev/yylo) | Command-line orchestrator for coding agents with typed task, validation, merge, and release-readiness boundaries, a dedicated branch/worktree, and risk-based merge review |
• SandBase Harness - harness) | Local-first, self-hosted TypeScript agent runtime and MCP bridge with sandboxed sessions and audit/replay |
• Tale - project/tale) | Self-hosted workspace for people and AI agents, with shared project tasks, manager delegation, persistent sandbox workspaces, and human review of results |
• Raven - AI/Raven) | Plans a complex task as a DAG and runs it across built-in and third-party agents, with a self-evolving loop that installs only verified changes |
• Markus - global/markus) | Platform for building and running teams of AI agents with roles, task delegation, and persistent memory |

📚 Research & Benchmarks

📊 Benchmarks

• ToolBench
• SOTOPIA-π - lab/sotopia-pi) | Social intelligence benchmark for multi-agent systems |
• PerspectiveGap - agent systems, 110 scenarios across 10 topologies |
• ClawBench - AI-Lab/ClawBench) | Live-web browser-agent benchmark with 283 tasks across 163 real websites |

📄 Papers

• arXiv - based agents |
• arXiv
• arXiv - agent systems |
• arXiv
• arXiv
• arXiv
• arXiv - agent pipeline with bandit scheduling turns optimization problems into solver code |
• arXiv - aware training for a live-commerce agent, with evaluation on changing skills, tools, prompts, and hooks |
• SSRN

🚀 Specialized Agents

💻 Coding Agents

• GPT Pilot - io/gpt-pilot) | Assists in writing and debugging code |
• Devika
• Plandex - ai/plandex) | AI coding engine for complex projects |
• Gemini CLI - gemini/gemini-cli) | Open-source AI agent bringing Gemini to the terminal with MCP support |
• AgenticSeek
• Cline
• bolt.diy - labs/bolt.diy) | AI-powered full-stack web development in the browser with 19+ LLMs |
• SWE-agent - agent/SWE-agent) | AI agent for software engineering tasks |
• OpenHands - source AI software development agents (formerly OpenDevin) |
• Aider - AI/aider) | AI pair programming in terminal |
• Goose - goose/goose) | Open-source extensible AI agent by Block for engineering tasks |
• Atomic Agent - ai/atomic-agent) | Local-first CLI and TUI coding agent, open-weight models, 56 tools, MCP |
• Orbi - build/orbi) | Takes a labeled GitHub issue to a reviewed PR and a tagged release |

🎨 Creative Agents

• ShortGPT - form video generation agent |
• AI-town - infra/ai-town) | Virtual world simulation with AI agents |
• BulkPublish - api) | API and agent skills for drafting, scheduling, and publishing social media content |

🗣️ Programming Language Agents

• TypeChat - safe LLM outputs using TypeScript types |
• LangChain.js - ai/langchainjs) | JS version of LangChain |
• Semantic Kernel - kernel) | Integrate LLMs into .NET apps |
• LangChain4j
• Haystack - ai/haystack) | Search and question answering agents |
• LangChainGo
• OpenAI Agents (Python) - agents-python) | Lightweight, provider-agnostic agent framework |
• Swarms-rs - Swarm-Corporation/swarms-rs) | Swarm-based agent orchestration in Rust |
• AnythingLLM - Labs/anything-llm) | All-in-one AI agent builder with RAG and UI |
• LangChain.rb - ai-core/langchainrb) | Ruby implementation of LangChain |

🔬 Research Agents

• GPT Researcher - researcher) | Autonomous agent for comprehensive research |
• Storm - oval/storm) | Multi-agent system for collaborative reasoning |
• DeerFlow - flow) | Framework for deep research with web search and Python execution |
• Agon - Factory/Agon) | Prompt Economy orchestrator: reusable scientist/coder/auditor loops instead of per-task prompts, 18 roles, 10+ disciplines |
• AutoNumerics - language problem descriptions |
• Dr. Claw - claw) | Local research workspace with survey, ideation, experiment, publication, and promotion stages |
• Caesar - agent) | Builds a knowledge graph while exploring the web, then refines drafts via adversarial synthesis |
• P2PCLAW - P2P) | Decentralized P2P agent swarm for open scientific research |
• CAJAL

🎙️ Voice Agents

• Pipecat - ai/pipecat) | Open-source framework for building real-time voice and multimodal conversational agents |
• LiveKit Agents - time voice AI agents with WebRTC transport, built on LiveKit |
• TEN Framework - framework/ten-framework) | Open-source framework for real-time conversational voice agents with multimodal support |

🌐 Web & Computer Use Agents

• NanoBrowser - based AI browsing agent |
• Browser Use - use/browser-use) | Make websites accessible for AI agents, automate tasks with ease |
• Firecrawl - ready markdown or structured data |
• Crawl4AI - source LLM-friendly web crawler and scraper |
• Stagehand
• Xquik - dev/x-twitter-scraper) | X/Twitter data and action API (REST, MCP, webhooks) that agents call for search, monitoring, and posting |
• Skyvern - AI/skyvern) | Browser automation agent using LLMs and computer vision to complete web workflows |
• Playwright MCP - mcp) | MCP server exposing Playwright browser automation to AI agents |
• Agent S - ai/Agent-S) | Open framework for computer-use agents that operate desktop GUIs like a human |
• UI-TARS Desktop - TARS-desktop) | GUI agent by ByteDance that controls desktop and browser via natural language, built on the UI-TARS vision model |
• agent-qa - qa) | Self-improving QA agent for natural-language web and mobile tests |