{"id":27004201,"url":"https://github.com/inclusionai/aworld","last_synced_at":"2026-02-03T06:13:00.268Z","repository":{"id":284572969,"uuid":"948379879","full_name":"inclusionAI/AWorld","owner":"inclusionAI","description":"Build, evaluate and run General Multi-Agent Assistance with ease","archived":false,"fork":false,"pushed_at":"2025-04-03T03:48:08.000Z","size":96017,"stargazers_count":134,"open_issues_count":3,"forks_count":3,"subscribers_count":9,"default_branch":"main","last_synced_at":"2025-04-03T04:28:13.787Z","etag":null,"topics":["agent-swarm","agentic-ai","computer-use","gym-environment","mcp","mcp-server","phone-use","world-model"],"latest_commit_sha":null,"homepage":"","language":"Python","has_issues":true,"has_wiki":null,"has_pages":null,"mirror_url":null,"source_name":null,"license":"mit","status":null,"scm":"git","pull_requests_enabled":true,"icon_url":"https://github.com/inclusionAI.png","metadata":{"files":{"readme":"README.md","changelog":null,"contributing":null,"funding":null,"license":"LICENSE","code_of_conduct":null,"threat_model":null,"audit":null,"citation":null,"codeowners":null,"security":null,"support":null,"governance":null,"roadmap":null,"authors":null,"dei":null,"publiccode":null,"codemeta":null}},"created_at":"2025-03-14T08:30:52.000Z","updated_at":"2025-04-02T12:26:41.000Z","dependencies_parsed_at":"2025-03-26T15:55:41.146Z","dependency_job_id":null,"html_url":"https://github.com/inclusionAI/AWorld","commit_stats":null,"previous_names":["inclusionai/aworld"],"tags_count":0,"template":false,"template_full_name":null,"repository_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/repositories/inclusionAI%2FAWorld","tags_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/repositories/inclusionAI%2FAWorld/tags","releases_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/repositories/inclusionAI%2FAWorld/releases","manifests_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/repositories/inclusionAI%2FAWorld/manifests","owner_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/owners/inclusionAI","download_url":"https://codeload.github.com/inclusionAI/AWorld/tar.gz/refs/heads/main","host":{"name":"GitHub","url":"https://github.com","kind":"github","repositories_count":247128753,"owners_count":20888235,"icon_url":"https://github.com/github.png","version":null,"created_at":"2022-05-30T11:31:42.601Z","updated_at":"2022-07-04T15:15:14.044Z","host_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub","repositories_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/repositories","repository_names_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/repository_names","owners_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/owners"}},"keywords":["agent-swarm","agentic-ai","computer-use","gym-environment","mcp","mcp-server","phone-use","world-model"],"created_at":"2025-04-04T06:15:18.811Z","updated_at":"2026-02-03T06:13:00.260Z","avatar_url":"https://github.com/inclusionAI.png","language":"Python","funding_links":[],"categories":[],"sub_categories":[],"readme":"\u003cdiv align=\"center\"\u003e\n\n# AWorld: Rich Environments, Intelligent Agents, Continuous Evolution\n\n\u003c/div\u003e\n\n\u003ch4 align=\"center\"\u003e\n\n*\"Self-awareness: the hardest problem isn't solving within limits, it's discovering one's own limitations\"*\n\n[![Twitter Follow][twitter-image]][twitter-url]\n[![WeChat QR Code][wechat-image]][wechat-url]\n[![Discord][discord-image]][discord-url]\n[![License: MIT][license-image]][license-url]\n[![DeepWiki][deepwiki-image]][deepwiki-url]\n[![arXiv][arxiv-image]][arxiv-url]\n[![Tutorial][tutorial-image]][tutorial-url]\n[![Playground][playground-image]][playground-url]\n\n\u003c/h4\u003e\n\n\u003ch4 align=\"center\"\u003e\n\n[中文版](./README_zh.md) |\n[Installation](#installation) |\n[Environments](#online-access-to-complex-environments) |\n[Agent](#efficient-agent-construction) |\n[Experience](#experience-to-samples) |\n[Training](#training) |\n[Architecture](#architecture-design-principles) |\n[Evolution](#evolution) |\n[Contributing](#contributing) |\n\n\u003c/h4\u003e\n\n---\n\n**AWorld (Agent World)** builds intelligent agents and rich environments where they operate, pushing the frontiers of AI capabilities and enabling continuous evolution. This project provides the fundamental recipe for agentic learning: [Environment Access](#online-access-to-complex-environments), [Agent Construction](#efficient-agent-construction), [Experience Retrieval](#experience-to-samples), and [Model Training](#training). What makes AWorld powerful is that agents can use these same components to automatically improve themselves.\n\n![](./readme_assets/aworld_loop.png)\n\n\u003e 💡 Visit our [homepage](https://www.aworldagents.com/) for more details, or try our online [environments](https://www.aworldagents.com/environments) and [agents](https://playground.aworldagents.com/). \n\n\n# Installation\n\u003e [!TIP]\n\u003e Python\u003e=3.11\n```bash\ngit clone https://github.com/inclusionAI/AWorld \u0026\u0026 cd AWorld\n\npip install -e .\n```\n\n# Online Access to Complex Environments\nProvisioning rich environments is hard—packages conflict, APIs need keys, concurrency must scale. We make it painless with three access modes:\n1. Use our default hosted setup (tooling with usage costs includes a limited free tier).\n2. Bring your own API keys for unrestricted access (coming soon).\n3. Pull our Docker images and run everything on your own infrastructure (coming soon).\n\n```python\nimport os\nimport asyncio\nfrom aworld.sandbox import Sandbox\n\nINVITATION_CODE = os.environ.get(\"INVITATION_CODE\", \"\")\n\nmcp_config = {\n    \"mcpServers\": {\n        \"gaia_server\": {\n            \"type\": \"streamable-http\",\n            \"url\": \"https://playground.aworldagents.com/environments/mcp\",\n            \"timeout\": 600,\n            \"sse_read_timeout\": 600,\n            \"headers\": {\n                \"ENV_CODE\": \"gaia\",\n                \"Authorization\": f\"Bearer {INVITATION_CODE}\",\n            }\n        }\n    }\n}\n\nasync def _list_tools():\n    sand_box = Sandbox(mcp_config=mcp_config, mcp_servers=[\"gaia_server\"])\n    return await sand_box.mcpservers.list_tools()\n\nif __name__ == \"__main__\":\n    tools = asyncio.run(_list_tools())\n    print(tools)\n```\n\n![](./readme_assets/how_to_access_env.gif)\n\n# Efficient Agent Construction\nIn Aworld, an agent is simply a model enhanced with tools. To spin one up, you only need:\n1. a model endpoint (for training, a vLLM service works great)\n2. an online environment to call (use our hosted options or plug in your own MCP toolchain)\nThat’s it—no heavyweight scaffolding required.\n\n```python\nfrom aworld.agents.llm_agent import Agent\nfrom aworld.runner import Runners\n\n# refer the section above for details\nmcp_config = {...}\n\nsearcher = Agent(\n    name=\"Search Agent\",\n    system_prompt=\"You specialize at searching.\",\n    mcp_config=mcp_config\n)\n\nif __name__ == \"__main__\":\n    result = Runners.sync_run(\n        input=\"Use google search tool to answer the question: the news about AI today.\",\n        agent=searcher\n    )\n    print(f\"answer: {result.answer}\")\n```\n\nRemember to plug in your LLM credentials first.\n```bash\n# Set LLM credentials\nexport LLM_MODEL_NAME=\"gpt-4\"\nexport LLM_API_KEY=\"your-api-key-here\"\nexport LLM_BASE_URL=\"https://api.openai.com/v1\"\n```\n\n## Complex Agent System Construction\n\nReal-world problems often need more than a single agent. AWorld gives you flexible build paths:\n1. design automated workflows end to end  [Docs](https://inclusionai.github.io/AWorld/Quickstart/workflow_construction/)\n2. spin up MCP-enabled agents [Docs](https://inclusionai.github.io/AWorld/Quickstart/agent_construction/)\n3. orchestrate multi-agent systems (MAS) [Docs](https://inclusionai.github.io/AWorld/Quickstart/multi-agent_system_construction/)\n\nWant to see it live? Load a pre-built DeepResearch team in the AWorld [Playground](https://playground.aworldagents.com/), inspect the source, and run it end to end.\n![](./readme_assets/playground_gaiateam.gif)\n\n# Experience to samples\nOur runtime captures every step across offline and online runs. Each task yields a complete trajectory—every LLM call, action, and reward—so you can synthesize training samples, audit performance, and iterate with confidence.\n\n## Complete Task Trajectories\nTasks unfold over many LLM calls. The framework captures every step, giving you a full trajectory.\n\n```python\nimport asyncio\nfrom aworld.runner import Runners\nfrom aworld.core.task import Task\nfrom aworld.logs.util import logger\nimport json\n\n# refer the section above for agent constrution \nsearcher = Agent(...)\n\nif __name__ == \"__main__\":\n    async def test_complete_trajectory():\n        task = Task(\n            input=\"Use google search tool to answer the question: the news about AI today.\",\n            agent=searcher\n        )\n\n        responses = await Runners.run_task(task)\n        resp = responses[task.id]\n        logger.info(f\"task answer: {resp.answer}\")\n        logger.info(f\"task trajectory: {json.dumps(resp.trajectory, ensure_ascii=False)}\")\n    asyncio.run(test_complete_trajectory())\n```\n\n## Single-Step Introspection\nNeed finer control? Call `step()` to inspect one action/response pair at a time. This lets you inject intermediate rewards during training, enabling richer, more flexible learning signals.\n\n```python\nimport os\nimport asyncio\nfrom aworld.runner import Runners\nfrom aworld.core.task import Task\nfrom aworld.logs.util import logger\nimport json\nfrom aworld.config import TaskConfig, TaskRunMode\n\n# refer the section above for agent constrution \nsearcher = Agent(...)\n\nif __name__ == \"__main__\":\n    async def test_single_step_introspection():\n        task = Task(\n            input=\"Use google search tool to answer the question: the news about AI today.\",\n            agent=searcher,\n            conf=TaskConfig(\n                resp_carry_context=True,\n                run_mode=TaskRunMode.INTERACTIVE\n            )\n        )\n\n        trajectory_log = os.path.join(os.path.dirname(__file__), \"trajectory_log.txt\")\n        is_finished = False\n        step = 1\n        while not is_finished:\n            with open(trajectory_log, \"a\", encoding=\"utf-8\") as traj_file:\n                is_finished, observation, response = await Runners.step(task)\n                traj_file.write(f\"Step {step}\\n\")\n                traj_file.write(json.dumps(response.trajectory, ensure_ascii=False, indent=2))\n                traj_file.write(\"\\n\\n\")\n                step += 1\n    asyncio.run(test_single_step_introspection())\n```\n\n# Training\nOnce agents can roam across environments, AWorld closes the loop with two complementary training modes that drive continuous improvement.\n\n## Model Training\nPlug any mainstream LLM trainer—AReal, Swift, Verl, Slime, etc.—into the runtime to update model parameters directly. Adapters are lightweight, so you can reuse the same environment and agent code across trainers.\n\n```python\nfrom datasets import load_dataset\nfrom aworld.agents.llm_agent import Agent\nfrom aworld.config import AgentConfig\n\nfrom train.trainer.agent_trainer import AgentTrainer\nfrom train.examples.train_gaia_with_aworld_verl.metrics.gaia_reward_function import gaia_reward_func\n\n\n# refer the section above for details\nmcp_config = {...}\n\n# Configure agent to use Verl as the model service (adapts inference format automatically)\nagent_config = AgentConfig(\n    llm_provider=\"verl\"\n)\nsearcher = Agent(\n    name=\"Search Agent\",\n    system_prompt=\"You specialize at searching.\",\n    mcp_config=mcp_config,\n    conf=agent_config\n)\n\ntrain_dataset = load_dataset(\"\", split=\"train\")\ntest_dataset = load_dataset(\"\", split=\"test\")\n\ntrainer = AgentTrainer(\n    agent=agent,\n    config=custom_train_config,\n    reward_func=gaia_reward_func,\n    train_dataset=train_dataset,\n    test_dataset=test_dataset\n)\n\ntrainer.train()\n```\n\u003e 💡 Check the [real case](./train/examples/train_gaia_with_aworld_verl/main.py) which includes the full training config to run agentic training.\n\n## Meta-Learning\nBeyond weights, you can meta-learn whole agent systems. Spin up role-specific agents that critique, rewrite prompts, refine workflow, or adjust strategies for a target agent, then iterate the team (e.g., our Gaia demo).\n\n![](./readme_assets/mas_meta_learning.png)\n\n# Architecture Design Principles\nThis framework is engineered to be highly adaptable, enabling researchers and developers to explore and innovate across multiple domains, thereby advancing the capabilities and applications of multi-agent systems.\n\n## Concepts \u0026 Framework\n| Concepts | Description |\n| :-------------------------------------- | ------------ |\n| [`agent`](./aworld/core/agent/base.py)  | Define the foundational classes, descriptions, output parsing, and multi-agent collaboration (swarm) logic for defining, managing, and orchestrating agents in the AWorld system. |\n| [`runner`](./aworld/runners)            | Contains runner classes that manage the execution loop for agents in environments, handling episode rollouts and parallel training/evaluation workflows.   |\n| [`task`](./aworld/core/task.py)         | Define the base Task class that encapsulates environment objectives, necessary tools, and termination conditions for agent interactions.  |\n| [`swarm`](./aworld/core/agent/swarm.py) | Implement the SwarmAgent class managing multi-agent coordination and emergent group behaviors through decentralized policies. |\n| [`sandbox`](./aworld/sandbox)           | Provide a controlled runtime with configurable scenarios for rapid prototyping and validation of agent behaviors. |\n| [`tools`](./aworld/tools)               | Offer a flexible framework for defining, adapting, and executing tools for agent-environment interaction in the AWorld system. |\n| [`context`](./aworld/core/context)      | Feature a comprehensive context management system for AWorld agents, enabling complete state tracking, configuration management, prompt optimization, multi-task state handling, and dynamic prompt templating throughout the agent lifecycle.  |\n| [`memory`](./aworld/memory)             | Implement an extensible memory system for agents, supporting short-term and long-term memory, summarization, retrieval, embeddings, and integration.|\n| [`trace`](./aworld/trace)               | Feature an observable tracing framework for AWorld, enabling distributed tracing, context propagation, span management, and integration with popular frameworks and protocols to monitor and analyze agent, tool, and task execution.|\n\n\n## Characteristics\n| Agent Construction            | Topology Orchestration                                                                            | Environment                    |\n|:------------------------------|:--------------------------------------------------------------------------------------------------|:-------------------------------|\n| ✅ Integrated MCP services     | ✅ Encapsulated runtime                                                                            | ✅ Runtime state management  |\n| ✅ Multi-model providers       | ✅ Flexible MAS patterns                                                                           | ✅ High-concurrency support  |\n| ✅ Customization options       | ✅ Clear state tracing                                                                             | ✅ Distributed training      |\n| ✅ [Support Agent Skills](https://github.com/inclusionAI/AWorld/tree/main/examples/skill_agent)  | [Support Aworld-Cli](https://github.com/inclusionAI/AWorld/tree/main/examples/aworld_cli_demo) 🚀 |       |\n\n\n# Evolution\nOur mission: AWorld handles the complexity, you focus on innovation. This section showcases cutting-edge multi-agent systems built with AWorld, advancing toward AGI.\n\n#### Agent Benchmarking\n\n\u003ctable style=\"width: 100%; border-collapse: collapse; table-layout: fixed;\"\u003e\n  \u003cthead\u003e\n    \u003ctr\u003e\n      \u003cth style=\"width: 30%; text-align: left; border-bottom: 2px solid #ddd; padding: 8px;\"\u003eCategory\u003c/th\u003e\n      \u003cth style=\"width: 20%; text-align: left; border-bottom: 2px solid #ddd; padding: 8px;\"\u003eAchievement\u003c/th\u003e\n      \u003cth style=\"width: 20%; text-align: left; border-bottom: 2px solid #ddd; padding: 8px;\"\u003ePerformance\u003c/th\u003e\n      \u003cth style=\"width: 25%; text-align: left; border-bottom: 2px solid #ddd; padding: 8px;\"\u003eKey Innovation\u003c/th\u003e\n      \u003cth style=\"width: 5%; text-align: left; border-bottom: 2px solid #ddd; padding: 8px;\"\u003eDate\u003c/th\u003e\n    \u003c/tr\u003e\n  \u003c/thead\u003e\n  \u003ctbody\u003e\n    \u003ctr\u003e\n      \u003ctd style=\"padding: 8px; vertical-align: top;\"\u003e🤖 Agent\n        \u003cbr\u003e\n        \u003ca href=\"https://playground.aworldagents.com/\" target=\"_blank\" style=\"text-decoration: none;\"\u003e\n          \u003cimg src=\"https://img.shields.io/badge/Try-Online-9B59B6?style=flat-square\" alt=\"Try Online\"\u003e\n        \u003c/a\u003e\n      \u003c/td\u003e\n      \u003ctd style=\"padding: 8px; vertical-align: top;\"\u003e\n        \u003cstrong\u003eGAIA Benchmark \u003cbr\u003eExcellence\u003c/strong\u003e\n        \u003cbr\u003e\n        \u003ca href=\"https://huggingface.co/spaces/gaia-benchmark/leaderboard\" target=\"_blank\" style=\"text-decoration: none;\"\u003e\n          \u003cimg src=\"https://img.shields.io/badge/GAIA-Leaderboard-blue\" alt=\"GAIA\"\u003e\n        \u003c/a\u003e\n      \u003c/td\u003e\n      \u003ctd style=\"padding: 8px; vertical-align: top;\"\u003e\n        Pass@1: \u003cstrong\u003e67.89\u003c/strong\u003e \u003cbr\u003e\n        Pass@3: \u003cstrong\u003e83.49\u003c/strong\u003e\n        \u003cbr\u003e (109 tasks)\n        \u003ca href=\"./examples/gaia/README_GUARD.md\" target=\"_blank\" style=\"text-decoration: none;\"\u003e\n          \u003cimg src=\"https://img.shields.io/badge/Code-README-green\" alt=\"Code\"\u003e\n        \u003c/a\u003e\n      \u003c/td\u003e\n      \u003ctd style=\"padding: 8px; vertical-align: top;\"\u003e\n        Multi-agent system \u003cbr\u003estability \u0026 orchestration\n        \u003cbr\u003e\n        \u003ca href=\"https://arxiv.org/abs/2508.09889\" target=\"_blank\" style=\"text-decoration: none;\"\u003e\n          \u003cimg src=\"https://img.shields.io/badge/Paper-arXiv-red\" alt=\"Paper\"\u003e\n        \u003c/a\u003e\n      \u003c/td\u003e\n      \u003ctd style=\"padding: 8px; vertical-align: top;\"\u003e2025/08/06\u003c/td\u003e\n    \u003c/tr\u003e\n    \u003ctr\u003e\n      \u003ctd style=\"padding: 8px; vertical-align: top;\"\u003e🧠 Reasoning\u003c/td\u003e\n      \u003ctd style=\"padding: 8px; vertical-align: top;\"\u003e\n        \u003cstrong\u003eIMO 2025 \u003cbr\u003eProblem Solving\u003c/strong\u003e\n        \u003cbr\u003e\n        \u003ca href=\"https://www.imo-official.org/year_info.aspx?year=2025\" target=\"_blank\" style=\"text-decoration: none;\"\u003e\n          \u003cimg src=\"https://img.shields.io/badge/IMO-2025-blue\" alt=\"IMO\"\u003e\n        \u003c/a\u003e\n      \u003c/td\u003e\n      \u003ctd style=\"padding: 8px; vertical-align: top;\"\u003e\n        \u003cstrong\u003e5/6\u003c/strong\u003e problems \u003cbr\u003esolved in 6 hours\n        \u003cbr\u003e\n        \u003ca href=\"examples/imo/README.md\" target=\"_blank\" style=\"text-decoration: none;\"\u003e\n          \u003cimg src=\"https://img.shields.io/badge/Code-README-green\" alt=\"Code\"\u003e\n        \u003c/a\u003e\n      \u003c/td\u003e\n      \u003ctd style=\"padding: 8px; vertical-align: top;\"\u003eMulti-agent collaboration \u003cbr\u003ebeats solo models\u003c/td\u003e\n      \u003ctd style=\"padding: 8px; vertical-align: top;\"\u003e2025/07/25\u003c/td\u003e\n    \u003c/tr\u003e\n    \u003ctr\u003e\n      \u003ctd style=\"padding: 8px; vertical-align: top;\"\u003e🖼️ Multi-Modal\u003c/td\u003e\n      \u003ctd style=\"padding: 8px; vertical-align: top;\"\u003e\n        \u003cstrong\u003eOSWorld \u003cbr\u003eRank 1st\u003c/strong\u003e\n        \u003cbr\u003e\n        \u003ca href=\"https://os-world.github.io/\" target=\"_blank\" style=\"text-decoration: none;\"\u003e\n          \u003cimg src=\"https://img.shields.io/badge/OSWorld-Leaderboard-green\" alt=\"OSWorld\"\u003e\n        \u003c/a\u003e\n      \u003c/td\u003e\n      \u003ctd style=\"padding: 8px; vertical-align: top;\"\u003e\n        \u003cstrong\u003e58.0%\u003c/strong\u003e \u003cbr\u003e Success Rate\n        \u003cbr\u003e\n        \u003ca href=\"examples/osworld/README.md\" target=\"_blank\" style=\"text-decoration: none;\"\u003e\n          \u003cimg src=\"https://img.shields.io/badge/Code-README-green\" alt=\"Code\"\u003e\n        \u003c/a\u003e\n      \u003c/td\u003e\n      \u003ctd style=\"padding: 8px; vertical-align: top;\"\u003eThe more tools the better?\u003c/td\u003e\n      \u003ctd style=\"padding: 8px; vertical-align: top;\"\u003e2025/09/18\u003c/td\u003e\n    \u003c/tr\u003e\n    \u003ctr\u003e\n      \u003ctd style=\"padding: 8px; vertical-align: top;\"\u003e🖼️ Multi-Modal\u003c/td\u003e\n      \u003ctd style=\"padding: 8px; vertical-align: top;\"\u003e\n        \u003cstrong\u003eVisualWebArena Rank 1st in September\u003c/strong\u003e\n        \u003cbr\u003e\n        \u003ca href=\"https://docs.google.com/spreadsheets/d/1M801lEpBbKSNwP-vDBkC_pF7LdyGU1f_ufZb_NWNBZQ/edit?gid=2044883967#gid=2044883967\" target=\"_blank\" style=\"text-decoration: none;\"\u003e\n          \u003cimg src=\"https://img.shields.io/badge/VWA-Leaderboard-green\" alt=\"VWA\"\u003e\n        \u003c/a\u003e\n      \u003c/td\u003e\n      \u003ctd style=\"padding: 8px; vertical-align: top;\"\u003e\n        \u003cstrong\u003e36.5%\u003c/strong\u003e \u003cbr\u003e Success Rate\n        \u003cbr\u003e\n        \u003ca href=\"examples/visualwebarena/README.md\" target=\"_blank\" style=\"text-decoration: none;\"\u003e\n          \u003cimg src=\"https://img.shields.io/badge/Code-README-green\" alt=\"Code\"\u003e\n        \u003c/a\u003e\n      \u003c/td\u003e\n      \u003ctd style=\"padding: 8px; vertical-align: top;\"\u003eAutomated tool generation \u003cbr\u003e\n        \u003ca href=\"https://arxiv.org/pdf/2509.21072\" target=\"_blank\" style=\"text-decoration: none;\"\u003e\n          \u003cimg src=\"https://img.shields.io/badge/Paper-arXiv-red\" alt=\"Paper\"\u003e\u003c/td\u003e\n      \u003ctd style=\"padding: 8px; vertical-align: top;\"\u003e2025/09/25\u003c/td\u003e\n    \u003c/tr\u003e\n    \u003ctr\u003e\n      \u003ctd style=\"padding: 8px; vertical-align: top;\"\u003e🔍 Deep-Search\u003c/td\u003e\n      \u003ctd style=\"padding: 8px; vertical-align: top;\"\u003e\n        \u003cstrong\u003eXbench Excellence\u003c/strong\u003e\n        \u003cbr\u003e\n        \u003ca href=\"https://xbench.org/\" target=\"_blank\" style=\"text-decoration: none;\"\u003e\n          \u003cimg src=\"https://img.shields.io/badge/xbench-Leaderboard-green\" alt=\"xbench\"\u003e\n        \u003c/a\u003e\n      \u003c/td\u003e\n      \u003ctd style=\"padding: 8px; vertical-align: top;\"\u003e\n        Pass@1: 51 \u003cbr\u003e Pass@3: 61\n        \u003cbr\u003e\n        \u003ca href=\"examples/xbench/README.md\" target=\"_blank\" style=\"text-decoration: none;\"\u003e\n          \u003cimg src=\"https://img.shields.io/badge/Code-README-green\" alt=\"Code\"\u003e\n        \u003c/a\u003e\n      \u003c/td\u003e\n      \u003ctd style=\"padding: 8px; vertical-align: top;\"\u003e\n          AWorld has its own context engine: Amni.\n      \u003c/td\u003e\n      \u003ctd style=\"padding: 8px; vertical-align: top;\"\u003e2025/10/23\u003c/td\u003e\n    \u003c/tr\u003e\n  \u003c/tbody\u003e\n\u003c/table\u003e\n\n#### Data Synthesis\n\n1. **FunReason-MT Technical Report: Overcoming the Complexity Barrier in Multi-Turn Function Calling** arxiv, 2025. [paper](https://arxiv.org/abs/2510.24645), [code](https://github.com/inclusionAI/AWorld-RL), [model](https://huggingface.co/Bingguang/FunReason-MT), [dataset](https://huggingface.co/datasets/Bingguang/FunReason-MT)\n\n    *Zengzhuang Xu, Bingguang Hao, Zechuan Wang, Yuntao Wen, Maolin Wang, etc.*\n   \n2. **From Failure to Mastery: Generating Hard Samples for Tool-use Agents** arxiv, 2026. [paper](https://arxiv.org/abs/2601.01498), [code](https://github.com/inclusionAI/AWorld-RL), [model](https://huggingface.co/Bingguang/FunReason-MT), [dataset](https://huggingface.co/datasets/Bingguang/FunReason-MT)\n\n    *Bingguang Hao, Zengzhuang Xu, Yuntao Wen, Xinyi Xu, Yang Liu, etc.*\n\n\n#### Model Training\n\n1. **AWorld: Orchestrating the Training Recipe for Agentic AI.** arxiv, 2025. [paper](https://arxiv.org/abs/2508.20404), [code](https://github.com/inclusionAI/AWorld/tree/main/train), [model](https://huggingface.co/inclusionAI/Qwen3-32B-AWorld)\n\n    *Chengyue Yu, Siyuan Lu, Chenyi Zhuang, Dong Wang, Qintong Wu, etc.*\n\n2. **FunReason: Enhancing Large Language Models' Function Calling via Self-Refinement Multiscale Loss and Automated Data Refinement.** arxiv, 2025. [paper](https://arxiv.org/abs/2505.20192), [model](https://huggingface.co/Bingguang/FunReason)\n\n    *Bingguang Hao, Maolin Wang, Zengzhuang Xu, Cunyin Peng, etc.*\n\n3. **Exploring Superior Function Calls via Reinforcement Learning.** arxiv, 2025. [paper](https://arxiv.org/abs/2508.05118), [code](https://github.com/BingguangHao/RLFC)\n\n    *Bingguang Hao, Maolin Wang, Zengzhuang Xu, Yicheng Chen, etc.*\n\n4. **RAG-R1 : Incentivize the Search and Reasoning Capabilities of LLMs through Multi-query Parallelism.** arxiv, 2025. [paper](https://arxiv.org/abs/2507.02962), [code](https://github.com/inclusionAI/AgenticLearning), [model](https://huggingface.co/collections/endertzw/rag-r1-68481d7694b3fca8b809aa29)\n\n    *Zhiwen Tan, Jiaming Huang, Qintong Wu, Hongxuan Zhang, Chenyi Zhuang, Jinjie Gu*\n\n5. **V2P: From Background Suppression to Center Peaking for Robust GUI Grounding Task.** arxiv, 2025. [paper](https://arxiv.org/abs/2508.13634), [code](https://github.com/inclusionAI/AgenticLearning/tree/main/V2P)\n\n    *Jikai Chen, Long Chen, Dong Wang, Leilei Gan, Chenyi Zhuang, Jinjie Gu*\n\n6. **Don’t Just Fine-tune the Agent, Tune the Environment** arxiv, 2025. [paper](https://arxiv.org/abs/2510.10197)\n\n    *Siyuan Lu, Zechuan Wang, Hongxuan Zhang, Qintong Wu, Leilei Gan, Chenyi Zhuang, etc.*\n\n\n#### Meta Learning\n\n1. **Profile-Aware Maneuvering: A Dynamic Multi-Agent System for Robust GAIA Problem Solving by AWorld.** arxiv, 2025. [paper](https://arxiv.org/abs/2508.09889), [code](https://github.com/inclusionAI/AWorld/blob/main/examples/gaia/README_GUARD.md)\n\n    *Zhitian Xie, Qintong Wu, Chengyue Yu, Chenyi Zhuang, Jinjie Gu*\n\n2. **Recon-Act: A Self-Evolving Multi-Agent Browser-Use System via Web Reconnaissance, Tool Generation, and Task Execution.** arxiv, 2025. [paper](https://arxiv.org/pdf/2509.21072), [code](https://github.com/inclusionAI/AWorld/tree/main/examples/visualwebarena)\n\n    *Kaiwen He, Zhiwei Wang, Chenyi Zhuang, Jinjie Gu*\n\n\n# Contributing\nWe warmly welcome developers to join us in building and improving AWorld! Whether you're interested in enhancing the framework, fixing bugs, or adding new features, your contributions are valuable to us.\n\nFor academic citations or wish to contact us, please use the following BibTeX entry:\n\n```bibtex\n@misc{yu2025aworldorchestratingtrainingrecipe,\n      title={AWorld: Orchestrating the Training Recipe for Agentic AI}, \n      author={Chengyue Yu and Siyuan Lu and Chenyi Zhuang and Dong Wang and Qintong Wu and Zongyue Li and Runsheng Gan and Chunfeng Wang and Siqi Hou and Gaochi Huang and Wenlong Yan and Lifeng Hong and Aohui Xue and Yanfeng Wang and Jinjie Gu and David Tsai and Tao Lin},\n      year={2025},\n      eprint={2508.20404},\n      archivePrefix={arXiv},\n      primaryClass={cs.AI},\n      url={https://arxiv.org/abs/2508.20404}, \n}\n```\n\n# Star History\n![](https://api.star-history.com/svg?repos=inclusionAI/AWorld\u0026type=Date)\n\n\n\u003c!-- resource section start --\u003e\n\u003c!-- image links --\u003e\n[arxiv-image]: https://img.shields.io/badge/Paper-arXiv-B31B1B?style=for-the-badge\u0026logo=arxiv\u0026logoColor=white\n[blog-image]: https://img.shields.io/badge/Blog-Coming%20Soon-FF5722?style=for-the-badge\u0026logo=blogger\u0026logoColor=white\n[deepwiki-image]: https://img.shields.io/badge/DeepWiki-Explore-blueviolet?style=for-the-badge\u0026logo=wikipedia\u0026logoColor=white\n[discord-image]: https://img.shields.io/badge/Discord-Join%20us-blue?style=for-the-badge\u0026logo=discord\u0026logoColor=white\n[github-code-image]: https://img.shields.io/badge/Code-GitHub-181717?style=for-the-badge\u0026logo=github\u0026logoColor=white\n[huggingface-dataset-image]: https://img.shields.io/badge/Dataset-Coming%20Soon-007ACC?style=for-the-badge\u0026logo=dataset\u0026logoColor=white\n[huggingface-model-image]: https://img.shields.io/badge/Model-Hugging%20Face-FF6B6B?style=for-the-badge\u0026logo=huggingface\u0026logoColor=white\n[license-image]: https://img.shields.io/badge/License-MIT-yellow?style=for-the-badge\n[twitter-image]: https://img.shields.io/badge/Twitter-Follow%20us-1DA1F2?style=for-the-badge\u0026logo=twitter\u0026logoColor=white\n[wechat-image]: https://img.shields.io/badge/WeChat-Add%20us-green?style=for-the-badge\u0026logo=wechat\u0026logoColor=white\n[tutorial-image]: https://img.shields.io/badge/Tutorial-Get%20Started-FF6B35?style=for-the-badge\u0026logo=book\u0026logoColor=white\n[playground-image]: https://img.shields.io/badge/Playground-Try%20Online-9B59B6?style=for-the-badge\u0026logo=book\u0026logoColor=white\n\n\u003c!-- aworld links --\u003e\n[deepwiki-url]: https://deepwiki.com/inclusionAI/AWorld\n[discord-url]: https://discord.gg/b4Asj2ynMw\n[license-url]: https://opensource.org/licenses/MIT\n[twitter-url]: https://x.com/AWorldAgents\n[wechat-url]: https://raw.githubusercontent.com/inclusionAI/AWorld/main/readme_assets/aworld_wechat.png\n[arxiv-url]: https://arxiv.org/abs/2508.20404\n[tutorial-url]: https://inclusionai.github.io/AWorld/\n[playground-url]: https://playground.aworldagents.com/\n\n\u003c!-- funreason links --\u003e\n[funreason-code-url]: https://github.com/BingguangHao/FunReason\n[funreason-model-url]: https://huggingface.co/Bingguang/FunReason\n[funreason-paper-url]: https://arxiv.org/pdf/2505.20192\n\u003c!-- [funreason-dataset-url]: https://github.com/BingguangHao/FunReason --\u003e\n\u003c!-- [funreason-blog-url]: https://github.com/BingguangHao/FunReason --\u003e\n\n\u003c!-- deepsearch links --\u003e\n[deepsearch-code-url]: https://github.com/inclusionAI/AgenticLearning\n[deepsearch-dataset-url]: https://github.com/inclusionAI/AgenticLearning\n[deepsearch-model-url]: https://huggingface.co/collections/endertzw/rag-r1-68481d7694b3fca8b809aa29\n[deepsearch-paper-url]: https://arxiv.org/abs/2507.02962\n\n\u003c!-- badge --\u003e\n[MAS]: https://img.shields.io/badge/Mutli--Agent-System-EEE1CE\n[IMO]: https://img.shields.io/badge/IMO-299D8F\n[BFCL]: https://img.shields.io/badge/BFCL-8AB07D\n[GAIA]: https://img.shields.io/badge/GAIA-E66F51\n[Runtime]: https://img.shields.io/badge/AWorld-Runtime-287271\n[Leaderboard]: https://img.shields.io/badge/Leaderboard-FFE6B7\n[Benchmark]: https://img.shields.io/badge/Benchmark-FFE6B7\n[Cloud-Native]: https://img.shields.io/badge/Cloud--Native-B19CD7\n[Forward]: https://img.shields.io/badge/Forward-4A90E2\n[Backward]: https://img.shields.io/badge/Backward-7B68EE\n[Code]: https://img.shields.io/badge/Code-FF6B6B\n[Paper]: https://img.shields.io/badge/Paper-4ECDC4\n\n\n\u003c!-- resource section end --\u003e\n","project_url":"https://awesome.ecosyste.ms/api/v1/projects/github.com%2Finclusionai%2Faworld","html_url":"https://awesome.ecosyste.ms/projects/github.com%2Finclusionai%2Faworld","lists_url":"https://awesome.ecosyste.ms/api/v1/projects/github.com%2Finclusionai%2Faworld/lists"}