[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"project-72152":3},{"id":4,"name":5,"fullName":6,"owner":7,"repo":5,"description":8,"homepage":9,"htmlUrl":10,"language":11,"languages":10,"totalLinesOfCode":10,"stars":12,"forks":13,"watchers":14,"openIssues":15,"contributorsCount":16,"subscribersCount":16,"size":16,"stars1d":17,"stars7d":18,"stars30d":19,"stars90d":16,"forks30d":16,"starsTrendScore":20,"compositeScore":21,"rankGlobal":10,"rankLanguage":10,"license":22,"archived":23,"fork":23,"defaultBranch":24,"hasWiki":25,"hasPages":25,"topics":26,"createdAt":10,"pushedAt":10,"updatedAt":31,"readmeContent":32,"aiSummary":33,"trendingCount":16,"starSnapshotCount":16,"syncStatus":34,"lastSyncTime":35,"discoverSource":36},72152,"youtu-agent","TencentCloudADP\u002Fyoutu-agent","TencentCloudADP","A simple yet powerful agent framework that delivers with open-source models","https:\u002F\u002Ftencentcloudadp.github.io\u002Fyoutu-agent\u002F",null,"Python",4568,466,31,50,0,1,7,19,3,30.01,"Other",false,"main",true,[27,28,29,30],"agent-framework","agents","openai-agents","python","2026-06-12 02:02:59","# \u003Cimg src=\"docs\u002Fassets\u002Flogo.svg\" alt=\"Youtu-agent Logo\" height=\"24px\"> Youtu-Agent: Scaling Agent Productivity with Automated Generation and Hybrid Policy Optimization\n\n\u003Cdiv align=\"center\">\n\u003Ca href=\"https:\u002F\u002Ftencentcloudadp.github.io\u002Fyoutu-agent\u002F\">\u003Cimg src=https:\u002F\u002Fimg.shields.io\u002Fbadge\u002F📖-Documentation-blue.svg>\u003C\u002Fa>\n\u003Ca href=https:\u002F\u002Fgithub.com\u002FTencentCloudADP\u002Fyoutu-agent>\u003Cimg src=https:\u002F\u002Fimg.shields.io\u002Fbadge\u002FGitHub-Tencent-blue.svg>\u003C\u002Fa>\n\u003Ca href=https:\u002F\u002Fdeepwiki.com\u002FTencentCloudADP\u002Fyoutu-agent>\u003Cimg src=https:\u002F\u002Fimg.shields.io\u002Fbadge\u002FDeepWiki-Tencent-blue.svg>\u003C\u002Fa>\n\u003Ca href=https:\u002F\u002Farxiv.org\u002Fabs\u002F2512.24615>\u003Cimg src=https:\u002F\u002Fimg.shields.io\u002Fbadge\u002FarXiv-2512.24615-b31b1b.svg>\u003C\u002Fa>\n\u003Ca href=\"https:\u002F\u002Fhuggingface.co\u002Fcollections\u002Fyolay\u002Fyoutu-agent-rl\">\u003Cimg src=https:\u002F\u002Fimg.shields.io\u002Fbadge\u002F🤗%20HuggingFace-Youtu%20Agent%20RL-ffc107.svg>\u003C\u002Fa>\n\u003C\u002Fdiv>\n\n\u003Cp align=\"center\">\n| \u003Ca href=\"README_ZH.md\">\u003Cb>中文\u003C\u002Fb>\u003C\u002Fa>\n| \u003Ca href=\"README_JA.md\">\u003Cb>日本語\u003C\u002Fb>\u003C\u002Fa>\n| \u003Ca href=\"#-benchmark-performance\">\u003Cb>🌟 Performance\u003C\u002Fb>\u003C\u002Fa> \n| \u003Ca href=\"#-examples\">\u003Cb>💡 Examples\u003C\u002Fb> \u003C\u002Fa> \n| \u003Ca href=\"#-features\">\u003Cb>✨ Features\u003C\u002Fb> \u003C\u002Fa> \n| \u003Ca href=\"#-getting-started\">\u003Cb>🚀 Getting Started\u003C\u002Fb> \u003C\u002Fa> \n| 📢 \u003Ca href=\"https:\u002F\u002Fdiscord.gg\u002FQjqhkHQVVM\">\u003Cb>Join Discord\u003C\u002Fb>\u003C\u002Fa> or \u003Ca href=\"https:\u002F\u002Fgithub.com\u002Fuser-attachments\u002Fassets\u002F354cd8e7-e108-4348-9355-04440052f408\">\u003Cb>WeChat\u003C\u002Fb>\u003C\u002Fa> \n|\n\u003C\u002Fp>\n\n\n`Youtu-Agent` is a flexible, high-performance framework for building, running, and evaluating autonomous agents. Beyond topping the benchmarks, this framework delivers powerful agent capabilities, e.g. data analysis, file processing, and deep research, all with open-source models. Additionally, the framework supports experience-based learning and end-to-end RL training to enhance agent capabilities.\n\n\u003Cimg src=\"docs\u002Fassets\u002Fmascot.png\" alt=\"Youtu-agent Logo\" width=\"200\" align=\"left\" style=\"margin-right:20px;\">\n\nKey highlights:\n- **Verified performance**: Achieves state-of-the-art performance on WebWalkerQA (71.47%) and GAIA (72.8%) using purely open-weight models (e.g., `DeepSeek-V3`), establishing a strong open-source baseline.\n- **Automated Agent Generation**: Introduces two paradigms: a **Workflow** mode for standard tasks and a **Meta-Agent** mode for complex requirements. The framework supports automated generation of tool code, prompts, and configurations, achieving over 81% tool synthesis success rate.\n- **Continuous Experience Learning**: The **Agent Practice** module enables low-cost continuous evolution via [Training-Free GRPO](https:\u002F\u002Farxiv.org\u002Fabs\u002F2510.08191). Agents accumulate experience and improve performance (e.g., +5.4% on AIME 2025) through in-context optimization without parameter updates.\n- **Scalable and Stable Agent RL**: The **Agent RL** module provides a complete pipeline for end-to-end reinforcement learning. By integrating with distributed frameworks, it addresses stability and scalability challenges, achieving 40% training speedup and scaling to 128 GPUs.\n- **Open-source friendly & cost-aware**: Optimized for accessible, low-cost deployment without reliance on closed models.\n- **Practical use cases**: Out-of-the-box support for tasks like data analysis, literature review, personal file organization, retrieval-augmented generation, and PPT generation.\n- **Flexible architecture**: Built on [openai-agents](https:\u002F\u002Fgithub.com\u002Fopenai\u002Fopenai-agents-python), with extensible support for diverse model APIs (from `DeepSeek` to `gpt-oss`), tool integrations, and framework implementations.\n\n## 🗞️ News\n\n- 🎉 [2026-01-17] **Agent Skills now supported!** Extend your agents with modular, domain-specific knowledge and workflows inspired by Anthropic's Claude Code skills. [[documentation](https:\u002F\u002Ftencentcloudadp.github.io\u002Fyoutu-agent\u002Fhowto\u002Fskills\u002F)].\n- 🚀 [2026-01-04] **Youtu Tip & Youtu-LLM Released!** We are excited to introduce [**Youtu-Tip**](https:\u002F\u002Fyoutu-tip.com\u002F), an extension of Youtu-Agent that runs on macOS and is powered by offline models (via Ollama). It automates tasks like file reading and web browsing. In the future, you will be able to run your agent built with Youtu-Agent even more easily using Youtu-Tip. Also, try [**Youtu-LLM**](https:\u002F\u002Fgithub.com\u002FTencentCloudADP\u002Fyoutu-tip\u002Ftree\u002Fmaster\u002Fyoutu-llm) inside.\n- 🚀 [2025-12-10] **Youtu-Agent x Agent-Lightning training integration available!** We've collaborated with the [Agent-Lightning](https:\u002F\u002Fgithub.com\u002Fmicrosoft\u002Fagent-lightning\u002F) team to implement efficient model training in verious scenarios. With ours efforts, training can now seamlessly scale to multi-node deployment on 128 GPUs. See details in the [rl\u002Fagl branch](https:\u002F\u002Fgithub.com\u002FTencentCloudADP\u002Fyoutu-agent\u002Ftree\u002Frl\u002Fagl).\n- 🎉 [2025-11-12] **Training-Free GRPO now available in main branch!** The agent practice module powered by [Training-Free Group Relative Policy Optimization](https:\u002F\u002Farxiv.org\u002Fabs\u002F2510.08191) is now integrated into the main branch. Enhance your agents' performance without fine-tuning at minimal cost (~$8 for RL runs). See our [Agent Practice Documentation](https:\u002F\u002Ftencentcloudadp.github.io\u002Fyoutu-agent\u002Fpractice\u002F) for usage and examples on math reasoning and web search tasks.\n- 📢 [2025-11-03] New examples: we add the [**PPT generation**](examples\u002Fppt_gen\u002FREADME.md) and [**RAG**](configs\u002Fagents\u002Fexamples\u002Frag.yaml) examples.\n- 🚀 [2025-10-10] [**Training-Free Group Relative Policy Optimization**](https:\u002F\u002Farxiv.org\u002Fabs\u002F2510.08191). RL for DeepSeek-V3.2 at $8? Yes, it's possible! Training-free GRPO keeps DeepSeek-V3.2 frozen, learns a token prior from ~100 samples for ~$8 RL runs, delivers verified math and web search gains! [code in branch [training_free_GRPO](https:\u002F\u002Fgithub.com\u002FTencentCloudADP\u002Fyoutu-agent\u002Ftree\u002Ftraining_free_GRPO)] [[x thread](https:\u002F\u002Fx.com\u002Fcai_cecilia47\u002Fstatus\u002F1976558824640393559)].\n- 🛠️ [2025-09-28] Agent auto-generation now ships with companion tooling: describe a capability once and let `Youtu-Agent` build the tool for you. [[details](https:\u002F\u002Ftencentcloudadp.github.io\u002Fyoutu-agent\u002Fauto_generation\u002F)].\n\n\u003Cdetails>\n\u003Csummary>\u003Cb>📰 Previous announcements\u003C\u002Fb>\u003C\u002Fsummary>\n\n- 📺 [2025-09-09] We hosted a live sharing the design philosophy and basic usage of `Youtu-Agent`. [[video](https:\u002F\u002Fwww.bilibili.com\u002Fvideo\u002FBV1mypqz4EvS)] [[documentation](https:\u002F\u002Fdoc.weixin.qq.com\u002Fdoc\u002Fw3_AcMATAZtAPICNLgt3CbnxRWaYWnW4)].\n- 🎁 [2025-09-02] [Tencent Cloud International](https:\u002F\u002Fwww.tencentcloud.com\u002F) offers new users of the DeepSeek API **3 million free tokens** (**Sep 1 – Oct 31, 2025**). [Try it out](https:\u002F\u002Fwww.tencentcloud.com\u002Fdocument\u002Fproduct\u002F1255\u002F70381) for free if you want to use DeepSeek models in `Youtu-Agent`! For enterprise agent solutions, also check out [Agent Development Platform](https:\u002F\u002Fadp.tencentcloud.com) (ADP).\n- 📺 [2025-08-28] We hosted a live sharing updates about DeepSeek-V3.1 and how to use it in the `Youtu-Agent` framework. [[video](https:\u002F\u002Fwww.bilibili.com\u002Fvideo\u002FBV1XwayzrETi\u002F)] [[documentation](https:\u002F\u002Fdoc.weixin.qq.com\u002Fdoc\u002Fw3_AcMATAZtAPICNvcLaY5FvTOuo7MwF)].\n\n\u003C\u002Fdetails>\n\n## 🌟 Benchmark Performance\n\n`Youtu-Agent` is built on open-source models and lightweight tools, demonstrating strong results on challenging deep search and tool use benchmarks.\n\n- **[WebWalkerQA](https:\u002F\u002Fhuggingface.co\u002Fdatasets\u002Fcallanwu\u002FWebWalkerQA)**: Achieved 60.71% accuracy with `DeepSeek-V3-0324`， using new released `DeepSeek-V3.1` can further improve to 71.47%, setting a new SOTA performance.\n- **[GAIA](https:\u002F\u002Fgaia-benchmark-leaderboard.hf.space\u002F)**: Achieved 72.8% pass@1 on the [text-only validation subset](https:\u002F\u002Fgithub.com\u002Fsunnynexus\u002FWebThinker\u002Fblob\u002Fmain\u002Fdata\u002FGAIA\u002Fdev.json) using `DeepSeek-V3-0324` (including models used within tools). We are actively extending evaluation to the full GAIA benchmark with multimodal tools, and will release the trajectories in the near future. Stay tuned! ✨\n\n![WebWalkerQA](docs\u002Fassets\u002Fimages\u002Fbenchmark_webwalkerqa.png)\n\n## 💡 Examples\n\nClick on the images to view detailed videos.\n\n\u003Ctable>\n  \u003Ctr>\n    \u003Ctd style=\"border: 1px solid black; padding: 10px; width: 50%; vertical-align: top;\">\n      \u003Cstrong>Data Analysis\u003C\u002Fstrong>\u003Cbr>Analyzes a CSV file and generates an HTML report.\n    \u003C\u002Ftd>\n    \u003Ctd style=\"border: 1px solid black; padding: 10px; width: 50%; vertical-align: top;\">\n      \u003Cstrong>File Management\u003C\u002Fstrong>\u003Cbr>Renames and categorizes local files for the user.\n    \u003C\u002Ftd>\n  \u003C\u002Ftr>\n  \u003Ctr>\n    \u003Ctd style=\"border: 1px solid black; padding: 10px; width: 50%; vertical-align: top;\">\n      \u003Cvideo src=\"https:\u002F\u002Fgithub.com\u002Fuser-attachments\u002Fassets\u002F60193435-b89d-47d3-8153-5799d6ff2920\" \n             poster=\"https:\u002F\u002Fimg.youtube.com\u002Fvi\u002Fr9we4m1cB6M\u002Fsddefault.jpg\" \n             controls muted preload=\"metadata\" \n             width=\"100%\" height=\"300\"\n             style=\"object-fit: cover; border-radius: 8px;\">\u003C\u002Fvideo>\n    \u003C\u002Ftd>\n    \u003Ctd style=\"border: 1px solid black; padding: 10px; width: 50%; vertical-align: top;\">\n      \u003Cvideo src=\"https:\u002F\u002Fgithub.com\u002Fuser-attachments\u002Fassets\u002Fdbb9cfc6-3963-4264-ba93-9ba21c5a579e\" \n             poster=\"https:\u002F\u002Fimg.youtube.com\u002Fvi\u002FGdA4AapE2L4\u002Fsddefault.jpg\" \n             controls muted preload=\"metadata\" \n             width=\"100%\" height=\"300\"\n             style=\"object-fit: cover; border-radius: 8px;\">\u003C\u002Fvideo>\n    \u003C\u002Ftd>\n  \u003C\u002Ftr>\n  \u003Ctr >\n    \u003Ctd style=\"border: 1px solid black; padding: 10px; width: 50%; vertical-align: top;\">\n      \u003Cstrong>Wide Research\u003C\u002Fstrong>\u003Cbr>Gathers extensive information to generate a comprehensive report, replicating the functionality of Manus.\n    \u003C\u002Ftd>\n    \u003Ctd style=\"border: 1px solid black; padding: 10px; width: 50%; vertical-align: top;\">\n      \u003Cstrong>Paper Analysis\u003C\u002Fstrong>\u003Cbr>Parses a given paper, performs analysis, and compiles related literature to produce a final result.\n    \u003C\u002Ftd>\n  \u003C\u002Ftr>\n  \u003Ctr>\n    \u003Ctd style=\"border: 1px solid black; padding: 10px; width: 50%; vertical-align: top;\">\n      \u003Cvideo src=\"https:\u002F\u002Fgithub.com\u002Fuser-attachments\u002Fassets\u002F6fc75814-e565-4f94-9ab5-33e3e7788e92\" \n             poster=\"https:\u002F\u002Fimg.youtube.com\u002Fvi\u002Fv3QQg0WAnPs\u002Fsddefault.jpg\" \n             controls muted preload=\"metadata\" \n             width=\"100%\" height=\"300\"\n             style=\"object-fit: cover; border-radius: 8px;\">\u003C\u002Fvideo>\n    \u003C\u002Ftd>\n    \u003Ctd style=\"border: 1px solid black; padding: 10px; width: 50%; vertical-align: top;\">\n      \u003Cvideo src=\"https:\u002F\u002Fgithub.com\u002Fuser-attachments\u002Fassets\u002F09b24f94-30f0-4e88-9aaf-9f3bbf82e99d\" \n             poster=\"https:\u002F\u002Fimg.youtube.com\u002Fvi\u002FvBddCjjRk00\u002Fsddefault.jpg\" \n             controls muted preload=\"metadata\" \n             width=\"100%\" height=\"300\"\n             style=\"object-fit: cover; border-radius: 8px;\">\u003C\u002Fvideo>\n    \u003C\u002Ftd>\n  \u003C\u002Ftr>\n  \u003Ctr >\n    \u003Ctd style=\"border: 1px solid black; padding: 10px; width: 50%; vertical-align: top;\">\n      \u003Cstrong>RAG\u003C\u002Fstrong>\u003Cbr>A RAG example by integration with RAGFlow service.\n    \u003C\u002Ftd>\n    \u003Ctd style=\"border: 1px solid black; padding: 10px; width: 50%; vertical-align: top;\">\n      \u003Cstrong>PPT Generation\u003C\u002Fstrong>\u003Cbr>An example that generate PPT file according to given content.\n    \u003C\u002Ftd>\n  \u003C\u002Ftr>\n  \u003Ctr>\n    \u003Ctd style=\"border: 1px solid black; padding: 10px; width: 50%; vertical-align: top;\">\n      \u003Cvideo src=\"https:\u002F\u002Fgithub.com\u002Fuser-attachments\u002Fassets\u002F4d74ef6f-7a84-4102-9666-0fbfe02e0d2f\" \n             controls muted preload=\"metadata\" \n             width=\"100%\" height=\"300\"\n             style=\"object-fit: cover; border-radius: 8px;\">\u003C\u002Fvideo>\n    \u003C\u002Ftd>\n    \u003Ctd style=\"border: 1px solid black; padding: 10px; width: 50%; vertical-align: top;\">\n      \u003Cvideo src=\"https:\u002F\u002Fgithub.com\u002Fuser-attachments\u002Fassets\u002F91568e27-bf77-44d6-baa6-b178d2d88255\" \n             controls muted preload=\"metadata\" \n             width=\"100%\" height=\"300\"\n             style=\"object-fit: cover; border-radius: 8px;\">\u003C\u002Fvideo>\n    \u003C\u002Ftd>\n  \u003C\u002Ftr>\n\u003C\u002Ftable>\n\n> [!NOTE]\n> See the [`examples`](.\u002Fexamples) directory and [documentation](https:\u002F\u002Ftencentcloudadp.github.io\u002Fyoutu-agent\u002Fexamples\u002F) for more details.\n\n### 🤖 Automatic Tool and Agent Generation\n\nA standout feature of `Youtu-Agent` is its ability to **automatically generate tools alongside agent configurations**. Other frameworks often make you hand-code functions or hand-craft prompts before an agent can even run. Here, you simply describe the task: the built-in meta-agent interviews you, assembles the necessary tools, produces YAML configs, and saves everything so you can execute it immediately.\n\n```bash\n# Interactively clarify your requirements and auto-generate a config\npython scripts\u002Fgen_simple_agent.py\n\n# Run the generated config\npython scripts\u002Fcli_chat.py --config generated\u002Fxxx\n```\n\n\u003Ctable>\n  \u003Ctr>\n    \u003Ctd style=\"border: 1px solid black; padding: 10px; width: 50%; vertical-align: top;\">\n      \u003Cstrong>Automatic Agent Generation\u003C\u002Fstrong>\u003Cbr>Interactively clarify your requirements, automatically generate the agent configuration, and run it right away.\n    \u003C\u002Ftd>\n    \u003Ctd style=\"border: 1px solid black; padding: 10px; width: 50%; vertical-align: top;\">\n      \u003Cstrong>Automatic Tool Generation\u003C\u002Fstrong>\u003Cbr>Describe the behaviors you need, let the meta-agent draft tool code and schemas, then drop them straight into your workflow.\n    \u003C\u002Ftd>\n  \u003C\u002Ftr>\n  \u003Ctr>\n    \u003Ctd style=\"border: 1px solid black; padding:10px; vertical-align:top; width: 400px;\">\n      \u003Cvideo src=\"https:\u002F\u002Fgithub.com\u002Fuser-attachments\u002Fassets\u002F0c2ee833-507e-4141-8de4-148ff3d9f9ef\" \n             poster=\"https:\u002F\u002Fimg.youtube.com\u002Fvi\u002FJVpHDJtKBo8\u002Fmaxresdefault.jpg\" \n             controls muted preload=\"metadata\" \n             width=\"100%\" height=\"auto\" \n             style=\"object-fit: cover; border-radius: 8px;\">\u003C\u002Fvideo>\n    \u003C\u002Ftd>\n    \u003Ctd style=\"border: 1px solid black; padding:10px; vertical-align:top; width: 400px;\">\n      \u003Cvideo src=\"https:\u002F\u002Fgithub.com\u002Fuser-attachments\u002Fassets\u002F37878544-cfda-4a8a-9b42-a7361782c750\" \n             poster=\"https:\u002F\u002Fimg.youtube.com\u002Fvi\u002FzjGooBuqdSE\u002Fmaxresdefault.jpg\" \n             controls muted preload=\"metadata\" \n             width=\"100%\" height=\"auto\" \n             style=\"object-fit: cover; border-radius: 8px;\">\u003C\u002Fvideo>\n    \u003C\u002Ftd>\n  \u003C\u002Ftr>\n\u003C\u002Ftable>\n\n> [!NOTE]\n> See [documentation](https:\u002F\u002Ftencentcloudadp.github.io\u002Fyoutu-agent\u002Fauto_generation\u002F) for more details.\n\n## ✨ Features\n\n![features](docs\u002Fassets\u002Fimages\u002Fheader.png)\n\n### Design Philosophy\n- **Minimal design**: We try to keep the framework simple and easy to use, avoiding unnecessary overhead.\n- **Modular & configurable**: Flexible customization and easy integration of new components.\n- **Open-source model support & low-cost**: Promotes accessibility and cost-effectiveness for various applications.\n\n### Core Features\n- **Built on openai-agents**: Leveraging the foundation of [openai-agents](https:\u002F\u002Fgithub.com\u002Fopenai\u002Fopenai-agents-python) SDK, our framework inherits streaming, tracing, and agent-loop capabilities, ensuring compatibility with both `responses` and `chat.completions` APIs for seamless adaptation to diverse models like [gpt-oss](https:\u002F\u002Fgithub.com\u002Fopenai\u002Fgpt-oss).\n- **Fully asynchronous**: Enables high-performance and efficient execution, especially beneficial for evaluating benchmarks.\n- **Tracing & analysis system**: Beyond OTEL, our `DBTracingProcessor` system provides in-depth analysis of tool calls and agent trajectories. (will be released soon)\n\n### Automation\n- **YAML based configuration**: Structured and easily manageable agent configurations.\n- **Automatic agent generation**: Based on user requirements, agent configurations can be automatically generated.\n- **Tool generation & optimization**: Tool evaluation and automated optimization, and customized tool generation will be supported in the future.\n\n### Use Cases\n- **Deep \u002F Wide research**: Covers common search-oriented tasks.\n- **Webpage generation**: Examples include generating web pages based on specific inputs.\n- **Trajectory collection**: Supports data collection for training and research purposes.\n\n\n## 🤔 Why Choose Youtu-Agent?\n\n`Youtu-Agent` is designed to provide significant value to different user groups:\n\n### For Agents Researchers & LLM Trainers\n- A **simple yet powerful baseline** that is stronger than basic ReAct, serving as an excellent starting point for model training and ablation studies.\n- **One-click evaluation scripts** to streamline the experimental process and ensure consistent benchmarking.\n\n### For Agent Application Developers\n- A **proven and portable scaffolding** for building real-world agent applications.\n- **Ease of Use**: Get started quickly with simple scripts and a rich set of built-in toolkits.\n- **Modular Design**: Key components like `Environment` and `ContextManager` are encapsulated yet highly customizable.\n\n### For AI & Agent Enthusiasts\n- **Practical Use Cases**: The `\u002Fexamples` directory includes tasks like deep research report generation, data analysis, and personal file organization.\n- **Simplicity & Debuggability**: A rich toolset and visual tracing tools make development and debugging intuitive and straightforward.\n\n\n## 🧩 Core Concepts\n\n- **Agent**: An LLM configured with specific prompts, tools, and an environment.\n- **Toolkit**: An encapsulated set of tools that an agent can use.\n- **Environment**: The world in which the agent operates (e.g., a browser, a shell).\n- **ContextManager**: A configurable module for managing the agent's context window.\n- **Benchmark**: An encapsulated workflow for a specific dataset, including preprocessing, rollout, and judging logic.\n\nFor more design and implementation details, please refer to our [technical documentation](https:\u002F\u002Ftencentcloudadp.github.io\u002Fyoutu-agent\u002F).\n\n## 🚀 Getting Started\n\nYoutu-Agent provides complete code and examples to help you get started quickly. Follow the steps below to run your first agent, or refer to [`docker\u002FREADME.md`](.\u002Fdocker\u002FREADME.md) for a streamlined Docker-based setup with interactive frontend.\n\n### Setup\n\n#### Source Code Deployment\n\n> [!NOTE]\n> The project requires Python 3.12+. We recommend using [uv](https:\u002F\u002Fgithub.com\u002Fastral-sh\u002Fuv) for dependency management.\n\nFirst, make sure Python and uv are installed.\n\nThen clone the repository and sync dependencies:\n\n```bash\ngit clone https:\u002F\u002Fgithub.com\u002FTencentCloudADP\u002Fyoutu-agent.git\ncd youtu-agent\nuv sync  # or, `make sync`\nsource .\u002F.venv\u002Fbin\u002Factivate\ncp .env.example .env  # NOTE: You should then config the necessary API keys.\n```\n\nAfter copying the `.env.example` file, you need to fill in the necessary keys in the `.env` file, e.g. LLM API keys. For example:\n\n```bash\n# llm requires OpenAI API format compatibility\n# setup your LLM config , ref https:\u002F\u002Fapi-docs.deepseek.com\u002F\nUTU_LLM_TYPE=chat.completions\nUTU_LLM_MODEL=deepseek-chat\nUTU_LLM_BASE_URL=https:\u002F\u002Fapi.deepseek.com\u002Fv1\nUTU_LLM_API_KEY=replace-to-your-api-key\n```\n\n> [Tencent Cloud International](https:\u002F\u002Fwww.tencentcloud.com\u002F) offers new users of the DeepSeek API **3 million free tokens** (**Sep 1 – Oct 31, 2025**). [Try it out](https:\u002F\u002Fwww.tencentcloud.com\u002Fdocument\u002Fproduct\u002F1255\u002F70381) for free. Once you’ve applied, replace the API key in the .env file below:\n\n```bash\n# llm\n# setup your LLM config , ref https:\u002F\u002Fwww.tencentcloud.com\u002Fdocument\u002Fproduct\u002F1255\u002F70381\nUTU_LLM_TYPE=chat.completions\nUTU_LLM_MODEL=deepseek-v3\nUTU_LLM_BASE_URL=https:\u002F\u002Fapi.lkeap.cloud.tencent.com\u002Fv1\nUTU_LLM_API_KEY=replace-with-your-api-key\n```\n\n#### Docker Deployment\n\nPlease refer to [`docker\u002FREADME.md`](.\u002Fdocker\u002FREADME.md) for a streamlined Docker-based setup with interactive frontend.\n\n### Quick Start\n\nYoutu-agent ships with built-in configurations. For example, the config `configs\u002Fagents\u002Fsimple\u002Fbase_search.yaml` defines a simple agent equipped with a search tool:\n\n```yaml\ndefaults:\n  - \u002Fmodel\u002Fbase\n  - \u002Ftools\u002Fsearch@toolkits.search\n  - _self_\n\nagent:\n  name: simple-tool-agent\n  instructions: \"You are a helpful assistant that can search the web.\"\n```\n\nYou can launch an interactive CLI chatbot with this agent by running:\n\n```bash\n# NOTE: You need to set `SERPER_API_KEY` and `JINA_API_KEY` in `.env` for web search access.\n# (We plan to replace these with free alternatives in the future)\npython scripts\u002Fcli_chat.py --config simple\u002Fbase_search\n# To avoid using the search toolkit, you can run:\npython scripts\u002Fcli_chat.py --config simple\u002Fbase\n```\n\n📖 More details: [Quickstart Documentation](https:\u002F\u002Ftencentcloudadp.github.io\u002Fyoutu-agent\u002Fquickstart)\n\n### Explore More Examples\n\nThe repository provides multiple ready-to-use examples. Some examples require the agent to have internet search capabilities, so you’ll need to configure the tool APIs in the `.env` file under the tools module:\n\n```bash\n# tools\n# serper api key, ref https:\u002F\u002Fserper.dev\u002Fplayground\nSERPER_API_KEY=\u003CAccess the URL in the comments to get the API Key>\n# jina api key, ref https:\u002F\u002Fjina.ai\u002Freader\nJINA_API_KEY=\u003CAccess the URL in the comments to get the API Key>\n```\n\nFor example, to enable the agent to automatically search online for information and generate an SVG image on the topic of “DeepSeek V3.1 New Features,” run the following command:\n\n```bash\npython examples\u002Fsvg_generator\u002Fmain.py\n```\n\nIf you want to visualize the agent’s runtime status using the web UI, download the frontend package from the Youtu-Agent releases and install it locally:\n\n```bash\n# Download the frontend package\ncurl -LO https:\u002F\u002Fgithub.com\u002FTencent\u002FYoutu-agent\u002Freleases\u002Fdownload\u002Ffrontend%2Fv0.2.0\u002Futu_agent_ui-0.2.0-py3-none-any.whl\n\n# Install the frontend package\nuv pip install utu_agent_ui-0.2.0-py3-none-any.whl\n```\n\nNext, run the web version of the SVG image generation command:\n\n```bash\npython examples\u002Fsvg_generator\u002Fmain_web.py\n```\n\nOnce the terminal shows the following message, the deployment is successful. You can access the project by clicking the local link:\n\n```bash\nServer started at http:\u002F\u002F127.0.0.1:8848\u002F\n```\n\n![svg_generator_ui](https:\u002F\u002Fgithub.com\u002Fuser-attachments\u002Fassets\u002F337d327f-91ee-434e-bbcf-297dd4b26c28)\n\nGiven a research topic, the agent will automatically search the web, collect relevant information, and output an SVG visualization.\n\n![svg_generator_result](https:\u002F\u002Fgithub.com\u002Fuser-attachments\u002Fassets\u002F41aa7348-5f02-4daa-b5b2-225e35d21067)\n\n📖 Learn more: [Examples Documentation](https:\u002F\u002Ftencentcloudadp.github.io\u002Fyoutu-agent\u002Fexamples)\n\n### Run Evaluations\n\nYoutu-Agent also supports benchmarking on standard datasets. For example, to evaluate on `WebWalkerQA`:\n\n```bash\n# Prepare dataset. This script will download and process WebWalkerQA dataset, and save it to DB.\npython scripts\u002Fdata\u002Fprocess_web_walker_qa.py\n\n# Run evaluation with config `ww.yaml` with your custom `exp_id`. We choose the sampled small dataset `WebWalkerQA_15` for quick evaluation.\n# NOTE: `JUDGE_LLM_TYPE, JUDGE_LLM_MODEL, JUDGE_LLM_BASE_URL, JUDGE_LLM_API_KEY` should be set in `.env`. Ref `.env.full`.\npython scripts\u002Frun_eval.py --config_name ww --exp_id \u003Cyour_exp_id> --dataset WebWalkerQA_15 --concurrency 5\n```\n\nResults are stored and can be further analyzed in the evaluation platform. See [Evaluation Analysis](.\u002Ffrontend\u002Fexp_analysis\u002FREADME.md).\n\n![eval_analysis_overview](https:\u002F\u002Fgithub.com\u002Fuser-attachments\u002Fassets\u002F4a285b9e-d096-437e-9b8e-e5bf6b1924b6)\n\n![eval_analysis_detail](https:\u002F\u002Fgithub.com\u002Fuser-attachments\u002Fassets\u002F4ede525a-5e16-4d88-9ebb-01a7dca3aaec)\n\n📖 Learn more: [Evaluation Documentation](https:\u002F\u002Ftencentcloudadp.github.io\u002Fyoutu-agent\u002Feval)\n\n## 📖 Dive Deeper\n\nAfter getting started, you can learn more about the framework and its capabilities through our full documentation:\n\n- 📖 **[Full Documentation](https:\u002F\u002Ftencentcloudadp.github.io\u002Fyoutu-agent\u002F)**: Explore the core concepts, architecture, and advanced features.\n- 🚀 **[Quickstart Guide](https:\u002F\u002Ftencentcloudadp.github.io\u002Fyoutu-agent\u002Fquickstart\u002F)**: A detailed guide to get you up and running.\n- ❓ **[FAQ](https:\u002F\u002Ftencentcloudadp.github.io\u002Fyoutu-agent\u002Ffaq)**: Find answers to common questions and issues.\n\n## 🙏 Acknowledgements\n\nThis project builds upon the excellent work of several open-source projects:\n- [openai-agents](https:\u002F\u002Fgithub.com\u002Fopenai\u002Fopenai-agents-python)\n- [mkdocs-material](https:\u002F\u002Fgithub.com\u002Fsquidfunk\u002Fmkdocs-material)\n- [model-context-protocol](https:\u002F\u002Fgithub.com\u002Fmodelcontextprotocol\u002Fpython-sdk)\n\n## 🙌 Contributing\n\nWe welcome contributions from the community! If you'd like to help improve Youtu-Agent, please read our [**Contributing Guidelines**](.\u002FCONTRIBUTING.md) to get started.\n\n## 📚 Citation\n\nIf you find this work useful, please consider citing:\n\n```bibtex\n@misc{youtu_agent,\n      title={Youtu-Agent: Scaling Agent Productivity with Automated Generation and Hybrid Policy Optimization}, \n      author={Yuchen Shi and Yuzheng Cai and Siqi Cai and Zihan Xu and Lichao Chen and Yulei Qin and Zhijian Zhou and Xiang Fei and Chaofan Qiu and Xiaoyu Tan and Gang Li and Zongyi Li and Haojia Lin and Guocan Cai and Yong Mao and Yunsheng Wu and Ke Li and Xing Sun},\n      year={2025},\n      eprint={2512.24615},\n      archivePrefix={arXiv},\n      primaryClass={cs.AI},\n      url={https:\u002F\u002Farxiv.org\u002Fabs\u002F2512.24615}, \n}\n@misc{training_free_grpo,\n      title={Training-Free Group Relative Policy Optimization}, \n      author={Tencent Youtu Lab},\n      year={2025},\n      eprint={2510.08191},\n      archivePrefix={arXiv},\n      primaryClass={cs.CL},\n      url={https:\u002F\u002Farxiv.org\u002Fabs\u002F2510.08191}, \n}\n```\n","Youtu-Agent 是一个用于构建、运行和评估自主代理的灵活且高性能框架。它支持数据分析、文件处理及深度研究等强大功能，全部基于开源模型。该框架引入了两种模式：针对标准任务的工作流模式和满足复杂需求的元代理模式，能够自动生成工具代码、提示词和配置，实现了超过81%的工具合成成功率。此外，Youtu-Agent 通过无参数更新的经验学习方式提升代理性能，并提供完整的端到端强化学习流水线，解决了稳定性和可扩展性问题，实现高达40%的训练加速并支持128个GPU的扩展。适用于需要高效开发和持续优化智能代理的应用场景。",2,"2026-06-11 03:40:35","high_star"]