{"id":33385959,"url":"https://github.com/modelscope/AgentEvolver","last_synced_at":"2025-11-28T02:00:58.268Z","repository":{"id":324149909,"uuid":"1095563347","full_name":"modelscope/AgentEvolver","owner":"modelscope","description":"AgentEvolver: Towards Efficient Self-Evolving Agent System","archived":false,"fork":false,"pushed_at":"2025-11-14T03:05:34.000Z","size":11144,"stargazers_count":22,"open_issues_count":1,"forks_count":3,"subscribers_count":0,"default_branch":"main","last_synced_at":"2025-11-14T05:23:30.626Z","etag":null,"topics":["agent","llm","reinforcement-learning","self-evolving"],"latest_commit_sha":null,"homepage":"https://modelscope.github.io/AgentEvolver/","language":"Python","has_issues":true,"has_wiki":null,"has_pages":null,"mirror_url":null,"source_name":null,"license":"apache-2.0","status":null,"scm":"git","pull_requests_enabled":true,"icon_url":"https://github.com/modelscope.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,"zenodo":null,"notice":null,"maintainers":null,"copyright":null,"agents":null,"dco":null,"cla":null}},"created_at":"2025-11-13T08:09:51.000Z","updated_at":"2025-11-14T04:53:23.000Z","dependencies_parsed_at":null,"dependency_job_id":null,"html_url":"https://github.com/modelscope/AgentEvolver","commit_stats":null,"previous_names":["modelscope/agentevolver"],"tags_count":null,"template":false,"template_full_name":null,"purl":"pkg:github/modelscope/AgentEvolver","repository_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/repositories/modelscope%2FAgentEvolver","tags_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/repositories/modelscope%2FAgentEvolver/tags","releases_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/repositories/modelscope%2FAgentEvolver/releases","manifests_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/repositories/modelscope%2FAgentEvolver/manifests","owner_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/owners/modelscope","download_url":"https://codeload.github.com/modelscope/AgentEvolver/tar.gz/refs/heads/main","sbom_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/repositories/modelscope%2FAgentEvolver/sbom","scorecard":null,"host":{"name":"GitHub","url":"https://github.com","kind":"github","repositories_count":286080680,"owners_count":27291100,"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","status":"online","status_checked_at":"2025-11-28T02:00:06.623Z","response_time":58,"last_error":null,"robots_txt_status":"success","robots_txt_updated_at":"2025-07-24T06:49:26.215Z","robots_txt_url":"https://github.com/robots.txt","online":true,"can_crawl_api":true,"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","llm","reinforcement-learning","self-evolving"],"created_at":"2025-11-23T05:00:42.435Z","updated_at":"2025-11-28T02:00:58.261Z","avatar_url":"https://github.com/modelscope.png","language":"Python","funding_links":[],"categories":["Python","Runtime"],"sub_categories":["Evolve Agent"],"readme":"\u003cp align=\"center\"\u003e\n \u003cimg src=\"docs/img/logo.png\" alt=\"AgentEvolver Logo\" width=\"70%\"\u003e\n\u003c/p\u003e\n\u003ch2 align=\"center\"\u003eAgentEvolver: Towards Efficient Self-Evolving Agent System\u003c/h2\u003e\n\n\u003c!-- --- --\u003e\n\n\u003cp align=\"center\"\u003e\n  \u003c!-- \u003ca href=\"https://arxiv.org/abs/0000\"\u003e\u003cimg src=\"https://img.shields.io/badge/cs.MA-0000-B31C1C?logo=arxiv\u0026logoColor=B31C1C\" alt=\"arxiv\"/\u003e\u003c/a\u003e --\u003e\n  \u003ca href=\"https://www.python.org/\"\u003e\u003cimg src=\"https://img.shields.io/badge/python-3.11+-blue\" alt=\"Python Version\"\u003e\u003c/a\u003e\n  \u003ca href=\"./LICENSE\"\u003e\u003cimg src=\"https://img.shields.io/badge/license-Apache--2.0-black\" alt=\"License\"\u003e\u003c/a\u003e\n  \u003ca href=\"https://modelscope.github.io/AgentEvolver/\"\u003e\u003cimg src=\"https://img.shields.io/badge/docs-online-blue?logo=markdown\" alt=\"Documentation\"\u003e\u003c/a\u003e\n  \u003ca href=\"https://arxiv.org/abs/2511.10395\"\u003e\u003cimg src=\"https://img.shields.io/badge/arXiv-2511.10395-b31b1b.svg\" alt=\"arXiv\"\u003e\u003c/a\u003e\n  \u003ca href=\"https://deepwiki.com/modelscope/AgentEvolver\"\u003e\u003cimg src=\"https://deepwiki.com/badge.svg\" alt=\"deepwiki\"\u003e\u003c/a\u003e\n  \u003ca href=\"https://github.com/modelscope/AgentEvolver\"\u003e\u003cimg src=\"https://img.shields.io/github/stars/modelscope/AgentEvolver?style=social\" alt=\"GitHub Stars\"\u003e\u003c/a\u003e\n\u003c/p\u003e\n\n\n\u003c!-- \u003cp align=\"center\"\u003e\n  \u003cstrong\u003eAgentEvolver: An Efficient Self-Evolving Agent System\u003c/strong\u003e\u003cbr\u003e\n\u003c/p\u003e --\u003e\n\n**AgentEvolver** is an end-to-end, self-evolving training framework that unifies self-questioning, self-navigating, and self-attributing into a cohesive system. It empowers agents to autonomously\nimprove their capabilities, aiming for efficient, cost-effective, and continuous capability evolution.\n\n\n## 📰 News\n\n- **[2025-11]** 📄 [The AgentEvolver Technical Report is now available](https://arxiv.org/abs/2511.10395), detailing the framework’s architecture, methodology, and key findings.\n- **[2025-11]** 🧩 AgentEvolver v1 has been released now!\n\n\n## ✨ Why AgentEvolver\n\n\n\n🧠 AgentEvolver provides three **Self-Evolving Mechanisms** from Environment to Policy:\n\n- **Automatic Task Generation (Self-Questioning)** – Explore the environment and autonomously create diverse tasks, eliminating costly manual dataset construction.\n- **Experience-guided Exploration (Self-Navigating)** – Summarize and reuse cross-task experience, guiding higher-quality rollouts and improving exploration efficiency.\n- **Attribution-based Credit Assignment (Self-Attributing)** – Process long trajectories to uncover the causal contribution of intermediate steps, enabling fine-grained and efficient policy optimization.\n\n\u003cp align=\"center\"\u003e\n \u003cimg src=\"docs/img/flowchart.png\" alt=\"AgentEvolver Flowchart\" width=\"80%\"\u003e\n\u003c/p\u003e\n\n\n\n\n## 🔧 Architecture Design\nAgentEvolver adopts a service-oriented dataflow architecture, seamlessly integrating environment sandboxes, LLMs, and experience management into modular services.\n\n\u003cp align=\"center\"\u003e\n \u003cimg src=\"docs/img/system.png\" alt=\"system framework\" width=\"80%\"\u003e\n\u003c/p\u003e\n\n\n- **Environment Compatibility** – Standardized interfaces for seamless integration with a wide range of external environments and tool APIs.\n- **Flexible Context Manager** – Built-in utilities for managing multi-turn contexts and complex interaction logic, supporting diverse deployment scenarios.\n- **Modular \u0026 Extensible Architecture** – Decoupled components allow easy customization, secondary development, and future algorithm upgrades.\n\n\n## 🌟 Benchmark Performance\n\nPerformance comparison on the AppWorld and BFCL-v3 benchmarks. AgentEvolver achieves superior results while using substantially fewer parameters than larger baseline models.\n\n\u003cp align=\"center\"\u003e\n \u003cimg src=\"docs/img/performance.png\" alt=\"Benchmark Performance\" width=\"80%\"\u003e\n\u003c/p\u003e\n\nPerformance on two benchmarks. Columns show avg@8 and best@8 for each benchmark, plus their averages (Avg.). All values are in percent (%). **Bolded numbers** highlight the best results.\n\n| **Model** | **Params** | **AppWorld** | | **BFCL v3** | | **Avg.** | |\n|:---|:---:|:---:|:---:|:---:|:---:|:---:|:---:|\n| | | avg@8 | best@8 | avg@8 | best@8 | avg@8 | best@8 |\n| Qwen2.5-7B | 7B | 1.8 | 5.6 | 29.8 | 42.4 | 15.8 | 24.0 |\n| +Questioning | 7B | 23.2 | 40.3 | 49.0 | 60.6 | 36.1 | 50.5 |\n| +Questioning\u0026Navigating | 7B | 26.3 | 43.1 | 53.3 | 61.0 | 39.8 | 52.1 |\n| +Questioning\u0026Attributing | 7B | 25.7 | 43.7 | 56.8 | 65.3 | 41.3 | 54.5 |\n| **AgentEvolver (overall)** | **7B** | **32.4** | **51.2** | **57.9** | **69.0** | **45.2** | **60.1** |\n| | | | | | | | |\n| Qwen2.5-14B | 14B | 18.0 | 31.4 | 41.6 | 54.1 | 29.8 | 42.8 |\n| +Questioning | 14B | 44.3 | 65.5 | 60.3 | 72.1 | 52.3 | 68.8 |\n| +Questioning\u0026Navigating | 14B | 45.4 | 65.3 | 62.8 | 74.5 | 54.1 | 69.9 |\n| +Questioning\u0026Attributing | 14B | 47.8 | 65.6 | 64.9 | 76.3 | 56.4 | 71.0 |\n| **AgentEvolver (overall)** | **14B** | **48.7** | **69.4** | **66.5** | **76.7** | **57.6** | **73.1** |\n\n\n## 🚀 Quick Start\n### Step 1. Basic Dependency Installation\n\nMake sure you have **conda** and **cuda toolkit** installed.\n\nThen, set up the training environment by running the script\n\n```bash\nbash install.sh\n```\n\n\n### Step 2. Setup Env-Service (Appworld as example)\nThe script below sets up an environment for appworld.\n\n```bash\ncd env_service/environments/appworld \u0026\u0026 bash setup.sh\n```\n\n### Step 3. Setup ReMe (Optional)\nSet up the ReMe for experience management by running the script:\n```bash\nbash external/reme/install_reme.sh\n```\nFor more detailed installation, please refer to [ReMe](https://github.com/agentscope-ai/ReMe).\n\n### Step 4. Begin Training! 🚀 🚀\nCopy the `example.env` file to `.env` and modify the parameters, including your **API key**, **conda path**.\n\nUsing AgentEvolver launcher to start environment, log dashboard and training process altogether.\n\n```bash\nconda activate agentevolver\n\n# option 1: minimal example without ReMe (using built-in datasets within environments)\npython launcher.py --conf examples/basic.yaml --with-appworld\n\n# option 2: full example with ReMe (questioning + navigating + attributing)\npython launcher.py --conf examples/overall.yaml --with-appworld --with-reme\n```\n\n## 🧩 Advanced Usage\n\n### 🔧 Manual Execution\n\nFor users requiring fine-grained control over the training pipeline, we provide standalone execution scripts: \n\n- `bash examples/run_basic.sh` - Execute basic RL pipeline with GRPO using built-in datasets within environments.\n- `bash examples/run_overall.sh` - Run the complete self-evolving AgentEvolver pipeline with fully customizable configurations.\n\nRefer to the  **[QuickStart](docs/tutorial/quick_start.md)** for detailed usage instructions and configuration parameters.\n\n### 📄 Documentation\n\nFor detailed usage and customization, please refer to the following guidelines:\n\n- **[Environment Service](docs/guidelines/env_service.md)** - Set up and manage environment instances, integrate custom environments\n- **[Task Manager](docs/guidelines/task_manager.md)** - Explore environments, generate synthetic tasks, and curate training data for agent evolution\n- **[Experience Manager](docs/guidelines/exp_manager.md)** - Configure experience pool management and self-navigating mechanisms\n- **[Advantage Processor](docs/guidelines/adv_processor.md)** - Implement self-attributing mechanisms with ADCA-GRPO for fine-grained credit assignment\n\nFor API documentation and more details, visit our [documentation site](docs/index.md).\n\n## 🔮 Upcoming\n- **Evolution in multi-agent scenarios** – Investigate autonomous co-evolution strategies for agents operating within shared, interactive environments.\n- **Cross-stage collaborative self-evolution** – Explore methods that couple questioning, navigating, and attributing into coordinated loops for mutual enhancement.\n\n\u003c!-- ## 🌟 Contact Us --\u003e\n\n## 🙏 Acknowledgements\nThis project builds upon the excellent work of several open-source projects:\n\n- [ReMe](https://github.com/agentscope-ai/ReMe) - for experience summarization and management;\n- [veRL](https://github.com/volcengine/verl) - for distributed RL training;\n- [mkdocs](https://github.com/mkdocs/mkdocs) - for documentation.\n\n## 📚 Citation\nIf you find this work useful, please consider citing:\n\n```bibtex\n@misc{AgentEvolver2025,\n  title         = {AgentEvolver: Towards Efficient Self-Evolving Agent System},\n  author        = {Yunpeng Zhai and Shuchang Tao and Cheng Chen and Anni Zou and Ziqian Chen and Qingxu Fu and Shinji Mai and Li Yu and Jiaji Deng and Zouying Cao and Zhaoyang Liu and Bolin Ding and Jingren Zhou},\n  year          = {2025},\n  eprint        = {2511.10395},\n  archivePrefix = {arXiv},\n  primaryClass  = {cs.LG},\n  url           = {https://arxiv.org/abs/2511.10395}\n}\n```\n\n\n## ✨ Star History\n\n[![Star History Chart](https://api.star-history.com/svg?repos=modelscope/AgentEvolver\u0026type=date\u0026legend=top-left)](https://www.star-history.com/#modelscope/AgentEvolver\u0026type=date\u0026legend=top-left)\n","project_url":"https://awesome.ecosyste.ms/api/v1/projects/github.com%2Fmodelscope%2FAgentEvolver","html_url":"https://awesome.ecosyste.ms/projects/github.com%2Fmodelscope%2FAgentEvolver","lists_url":"https://awesome.ecosyste.ms/api/v1/projects/github.com%2Fmodelscope%2FAgentEvolver/lists"}