[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"project-74241":3},{"id":4,"name":5,"fullName":6,"owner":5,"repo":5,"description":7,"homepage":8,"htmlUrl":9,"language":10,"languages":9,"totalLinesOfCode":9,"stars":11,"forks":12,"watchers":13,"openIssues":14,"contributorsCount":15,"subscribersCount":15,"size":15,"stars1d":16,"stars7d":17,"stars30d":18,"stars90d":15,"forks30d":15,"starsTrendScore":19,"compositeScore":20,"rankGlobal":9,"rankLanguage":9,"license":9,"archived":21,"fork":21,"defaultBranch":22,"hasWiki":21,"hasPages":21,"topics":23,"createdAt":9,"pushedAt":9,"updatedAt":24,"readmeContent":25,"aiSummary":26,"trendingCount":15,"starSnapshotCount":15,"syncStatus":27,"lastSyncTime":28,"discoverSource":29},74241,"Claw-AI-Lab","Claw-AI-Lab\u002FClaw-AI-Lab","One dashboard. An entire research team.","https:\u002F\u002Fclawailab.ai\u002F",null,"Python",1412,67,23,1,0,4,7,46,12,18.5,false,"preview-v1.1.0",[],"2026-06-12 02:03:24","\u003Cp align=\"center\">\n  \u003Cimg src=\"assets\u002Flogo_v1.png\" width=\"200\" alt=\"Claw AI Lab\">\n\u003C\u002Fp>\n\n\u003Ch2 align=\"center\">\u003Cb>Claw AI Lab: An Autonomous Multi-Agent Research Team\u003C\u002Fb>\u003C\u002Fh2>\n\n\u003C!-- \u003Cp align=\"center\">\n  \u003Cb>\u003Ci>One Command. A Complete Team.\u003C\u002Fi>\u003C\u002Fb>\n\u003C\u002Fp> -->\n\n\u003Cp align=\"center\">\n  \u003Ca href=\"https:\u002F\u002Fclawailab.ai\">\u003Cimg src=\"https:\u002F\u002Fimg.shields.io\u002Fbadge\u002FHomepage-clawailab.ai-E63946?logo=google-chrome&logoColor=white\" alt=\"Homepage\">\u003C\u002Fa>\n  \u003Ca href=\"LICENSE\">\u003Cimg src=\"https:\u002F\u002Fimg.shields.io\u002Fbadge\u002FLicense-MIT-yellow.svg\" alt=\"MIT License\">\u003C\u002Fa>\n  \u003Ca href=\"https:\u002F\u002Fpython.org\">\u003Cimg src=\"https:\u002F\u002Fimg.shields.io\u002Fbadge\u002FPython-3.11%2B-3776AB?logo=python&logoColor=white\" alt=\"Python 3.11+\">\u003C\u002Fa>\n  \u003Ca href=\"https:\u002F\u002Fnodejs.org\">\u003Cimg src=\"https:\u002F\u002Fimg.shields.io\u002Fbadge\u002FNode.js-18%2B-339933?logo=node.js&logoColor=white\" alt=\"Node.js 18+\">\u003C\u002Fa>\n  \u003Ca href=\"https:\u002F\u002Fgithub.com\u002FClaw-AI-Lab\u002FClaw-AI-Lab\">\u003Cimg src=\"https:\u002F\u002Fimg.shields.io\u002Fbadge\u002FGitHub-Claw--AI--Lab-6f42c1?logo=github&logoColor=white\" alt=\"GitHub\">\u003C\u002Fa>\n\u003C\u002Fp>\n\n---\n\n## 🔥 Updates\n\n- __[2026.04.02]__: Preview v1.1.0 — powered by **Claw-Code Harness**.\n- __[2026.03.25]__: Preview v1.0.0 - initial release.\n\n---\n\n\u003C!-- \u003Cdiv align=\"center\">\n  \u003Cvideo src=\"https:\u002F\u002Fgithub.com\u002Fuser-attachments\u002Fassets\u002F6660e4eb-f2a8-4c39-b47e-cbaf753eca7b\" controls width=\"100%\">\u003C\u002Fvideo>\n\u003C\u002Fdiv> -->\n\n## 🤔 What Is This?\n\n**Claw AI Lab** is a lab-native multi-agent research platform for interactive and scalable AI-driven science. It enables users to create a full AI research lab from a single prompt, with customizable roles, research directions, and collaborative workflows, rather than relying on a single-agent or fixed serial pipeline. Claw orchestrates multiple agents and projects in parallel through a FIFO-based scheduling framework, maximizing compute utilization while supporting cross-project knowledge sharing and mutual improvement. Crucially, the system keeps humans in the loop: users can intervene whenever needed, provide feedback under ambiguity, inject new ideas, and iteratively refine the research process through rollback and continuation. Combined with a simple UI that reduces everything to prompts and clicks, Claw transforms automated research into a more intuitive, steerable, and laboratory-like experience.\n\n\u003Cp align=\"center\">\u003Cb>We welcome contributions from the community to make this project better together!\u003C\u002Fb>\u003Cbr>\u003Cb>You are warmly invited to scroll to the bottom of the page to join our group for beta testing and discussion.\u003C\u002Fb>\u003C\u002Fp>\n\n---\n\n## 🖥️ Claw AI Lab Dashboard\n\n\u003Cp align=\"center\">\n  \u003Cimg src=\"assets\u002Fui.png\" width=\"860\" alt=\"Claw AI Lab UI\"\u002F>\n  \u003Cbr\u002F>\n  \u003Cb>Launch projects, monitor agents, and inspect every artifact — all from a single interface.\u003C\u002Fb>\u003Cbr>\n  \u003Csub>Real-time event stream · Multi-project overview · One-click rollback & resume · Artifact inspector\u003C\u002Fsub>\n\u003C\u002Fp>\n\n---\n\n## ✨ Key Features\n\n\u003Ctable>\n\u003Ctr>\u003Ctd>🖥️\u003C\u002Ftd>\u003Ctd>\u003Cb>Interactive UI\u003C\u002Fb>\u003C\u002Ftd>\u003Ctd>Real-time web dashboard with event stream, data shelf, and multi-project monitoring\u003C\u002Ftd>\u003C\u002Ftr>\n\u003Ctr>\u003Ctd>🧬\u003C\u002Ftd>\u003Ctd>\u003Cb>Claw Code Harness\u003C\u002Fb>\u003C\u002Ftd>\u003Ctd>Reads your local codebases, datasets & checkpoints — writes runnable code back to disk\u003C\u002Ftd>\u003C\u002Ftr>\n\u003Ctr>\u003Ctd>🔬\u003C\u002Ftd>\u003Ctd>\u003Cb>End-to-End Pipeline\u003C\u002Fb>\u003C\u002Ftd>\u003Ctd>One prompt → paper + code + figures + experiment logs, fully autonomous\u003C\u002Ftd>\u003C\u002Ftr>\n\u003Ctr>\u003Ctd>🤝\u003C\u002Ftd>\u003Ctd>\u003Cb>Three Research Modes\u003C\u002Fb>\u003C\u002Ftd>\u003Ctd>\u003Cb>Explore\u003C\u002Fb> · \u003Cb>Discussion\u003C\u002Fb> (multi-agent debate) · \u003Cb>Reproduce\u003C\u002Fb>\u003C\u002Ftd>\u003C\u002Ftr>\n\u003C!-- \u003Ctr>\u003Ctd>📄\u003C\u002Ftd>\u003Ctd>\u003Cb>PDF Reference Upload\u003C\u002Fb>\u003C\u002Ftd>\u003Ctd>Upload reference papers as PDF — the system extracts and cites them automatically\u003C\u002Ftd>\u003C\u002Ftr> -->\n\u003C\u002Ftable>\n\n---\n\n### 🏆 Generated Project Showcase\n\nEach project autonomously produces a full research deliverable: **Paper** · **Code** · **Figures** · **Experiment Logs**\n\n\u003Ctable width=\"100%\">\n\u003Ctr>\n\u003Ctd align=\"center\" width=\"50%\">\n\u003Ca href=\"assets\u002Fshowcase\u002Fshowcase-quantify-hallucination.md\">\u003Cb>OATH: Quantifying Video Hallucination via Occlusion Debt\u003C\u002Fb>\u003C\u002Fa>\u003Cbr>\n\u003Csub>Lab Explore · CV · Video Generation Evaluation\u003C\u002Fsub>\u003Cbr>\u003Cbr>\n\u003Ca href=\"assets\u002Fshowcase\u002Fshowcase-quantify-hallucination.md\">\u003Cimg src=\"assets\u002Fshowcase\u002Fquantifying-hallucination\u002Fstage-22\u002Ffigures\u002Foath_pipeline_overview.png\" width=\"380\">\u003C\u002Fa>\u003Cbr>\n\u003Csub>Best method achieves \u003Cb>0.1714\u003C\u002Fb> primary error vs CLIP-T baseline \u003Cb>0.2393\u003C\u002Fb> (↓28%)\u003C\u002Fsub>\n\u003C\u002Ftd>\n\u003Ctd align=\"center\" width=\"50%\">\n\u003Ca href=\"assets\u002Fshowcase\u002Fshowcase-reproduce-phycustom-on-flux.md\">\u003Cb>Reproducing PhyCustom on FLUX\u003C\u002Fb>\u003C\u002Fa>\u003Cbr>\n\u003Csub>Reproduce · Image Gen · Multi-Concept Customization\u003C\u002Fsub>\u003Cbr>\u003Cbr>\n\u003Ca href=\"assets\u002Fshowcase\u002Fshowcase-reproduce-phycustom-on-flux.md\">\u003Cimg src=\"assets\u002Fshowcase\u002Freproduce-phycustom\u002Fstage-22\u002Ffigures\u002Fmethod_flip_pipeline_overview.png\" width=\"380\">\u003C\u002Fa>\u003Cbr>\n\u003Csub>5 methods × 3 seeds = 15 runs; output-space decoupling edges at \u003Cb>0.2813\u003C\u002Fb>\u003C\u002Fsub>\n\u003C\u002Ftd>\n\u003C\u002Ftr>\n\u003C!-- \u003Ctr>\n\u003Ctd align=\"center\">\u003Cimg src=\"assets\u002Fshowcase\u002Fquantifying-hallucination\u002Fstage-22\u002Flatex_package\u002Ffigures\u002Ffig_main_comparison.png\" width=\"360\">\u003Cbr>\u003Csub>Main comparison across 19 conditions\u003C\u002Fsub>\u003C\u002Ftd>\n\u003Ctd align=\"center\">\u003Cimg src=\"assets\u002Fshowcase\u002Freproduce-phycustom\u002Fstage-22\u002Flatex_package\u002Ffigures\u002Ffig_main_comparison.png\" width=\"360\">\u003Cbr>\u003Csub>Main comparison across 5 methods\u003C\u002Fsub>\u003C\u002Ftd>\n\u003C\u002Ftr> -->\n\u003C\u002Ftable>\n\n---\n\n### 🏆 Discussion Mode Showcase\n\nMulti-agent discussion on: **\"What is the most deployable direction for Video Action Models in Embodied AI?\"**\n\n> **Agent A** — World Model + MPC (Model Predictive Control) is the most industrially stable path.\n>\n> **Agent B** — \"Train with video, infer with action\" is the most deployable policy paradigm.\n>\n> **Agent C** — Execution monitoring & SOP (Standard Operating Procedure) automation lands fastest as a product.\n\n**Consensus:** The most deployable form is not a single end-to-end model, but a **layered, modular system** — use video supervision during training to learn rich dynamics, output actions directly at inference for low latency, and layer planning\u002FMPC\u002Fsafety modules on top for closed-loop robustness and recovery.\n\n\u003Cdetails>\n\u003Csummary>\u003Cb>Top 3 Research Directions (ranked by deployability)\u003C\u002Fb>\u003C\u002Fsummary>\n\n| # | Direction | Deployability |\n| :---: | :--- | :--- |\n| 1 | **Layered Video-Action Stack** — video-action joint training + direct action inference + MPC safety | Highest — best balance of latency, interpretability & safety |\n| 2 | **Video-to-Plan \u002F SOP** — demo videos → step sequences & skill graphs for existing robots | High — smallest embodiment gap, clearest commercial path |\n| 3 | **Execution Monitor** — real-time step tracking, anomaly detection, re-planning triggers | High — fastest to production; critical for industrial reliability |\n\n\u003C\u002Fdetails>\n\n\u003Cdetails>\n\u003Csummary>\u003Cb>Key Contradictions Resolved\u003C\u002Fb>\u003C\u002Fsummary>\n\n| Debate | Resolution |\n| :--- | :--- |\n| World Model + MPC vs. Direct Action? | **Combine both** — world model for representation, direct action for control, MPC for safety |\n| Human video: valuable or too much gap? | **Pre-training yes**; direct low-level transfer not yet reliable |\n| Is monitoring a \"real\" action model? | Not the backbone, but **fastest to reach production value** |\n\n\u003C\u002Fdetails>\n\n**[→ Full Transcript](assets\u002Fshowcase\u002Fdiscussion_transcript.md)** · **[→ Consensus Synthesis](assets\u002Fshowcase\u002Fconsensus_synthesis.md)**\n\n---\n\n## 🚀 Quick Start\n\n### 1. Install\n\n```bash\ngit clone https:\u002F\u002Fgithub.com\u002FClaw-AI-Lab\u002FClaw-AI-Lab.git\ncd Claw-AI-Lab\n\n# Create python environment\nconda create -n clawailab python=3.11\nconda activate clawailab\n\n# Backend\ncd backend\u002Fagent\npip install -e \".[all]\"\npip install websockets\n\n# Frontend\ncd ..\u002F..\u002Ffrontend\nnpm install\ncd ..\n\n# ML dependencies\n# You can add more packages based on your research project\npip install torch torchvision diffusers transformers accelerate safetensors datasets \\\n            huggingface_hub opencv-python pandas matplotlib scikit-image scipy einops tqdm\n```\n\n### 2. Configure\n\nFill in following configurations in examples\u002Fconfig_template.yaml:\n```\nllm:\n  base_url: \"your-api-endpoint\"\n  api_key: \"your-api-key\"\n  primary_model: \"gpt-5.4\"\n  coding_model: \"gpt-5.4\"\n  image_model: \"gemini-3-pro-image-preview\"\n  fallback_models:\n    - \"qwen3.5-plus\"\n    - \"qwen-plus\"\n\nsandbox:\n  python_path: \"\u002Fabsolute\u002Fpath\u002Fto\u002Fclawailab\u002Fbin\u002Fpython\"  # Get this path by running: conda activate clawailab && which python\n```\n\n### 3. Run\n\n```bash\n.\u002Fstart.sh              # Start all services\n.\u002Fstart.sh stop         # Stop\n.\u002Fstart.sh restart      # Restart\n.\u002Fstart.sh status       # Status check\n.\u002Fstart.sh fresh        # Clean restart (reset all data)\n```\n\nOpen **http:\u002F\u002Flocalhost:5903\u002F** → Submit your research topic and let the agents work.\n\n---\n\n## 💡 Tips to Get the Best Results\n\n| # | Recommendation | Why |\n|---|---|---|\n| 1 | **Prepare local codebases, datasets & checkpoints** — enter their paths when submitting a project | Avoids download delays and network failures during runs |\n| 2 | **Use a strong coding model like GPT 5.4** | Significantly better code quality and fewer iteration cycles |\n| 3 | **Review the `IMPORTANT` fields in [Configuration Details](#️-configuration-details)** | Misconfigured API keys or resource limits are the #1 cause of failed runs |\n\n---\n\n## ⚙️ Configuration Details\n\nEvery field in `examples\u002Fconfig_template.yaml` explained. Fields marked **IMPORTANT** are the ones you almost always need to set.\n\n\u003Cdetails>\n\u003Csummary>Click to expand full reference\u003C\u002Fsummary>\n\n```yaml\n# === Project ===\nproject:\n  name: \"my-project\"              # Project identifier, used for directory naming and UI display\n  mode: \"full-auto\"               # Pipeline mode: \"full-auto\" runs all stages without human gates\n\n# === Research ===\nresearch:\n  topic: \"Your research topic\"    # The research topic or paper to reproduce (required)\n  domains:                        # Research domains for literature search scope\n    - \"deep-learning\"\n  daily_paper_count: 5            # Number of papers to retrieve per search query\n  quality_threshold: 3.0          # Minimum relevance score (1-5) for literature screening\n  reference_papers: []            # List of reference paper titles or arXiv IDs\n\n# === Notifications ===\nnotifications:\n  channel: \"console\"              # Notification channel: \"console\" | \"discord\" | \"slack\"\n  on_stage_start: true            # Notify when a stage begins\n  on_gate_required: true          # Notify when human approval is needed\n\n# === Knowledge Base ===\nknowledge_base:\n  backend: \"markdown\"             # Storage format: \"markdown\" | \"obsidian\"\n  root: \"docs\u002Fkb\"                 # Root directory for knowledge base files\n\n# === OpenClaw Bridge ===\nopenclaw_bridge:\n  use_message: false              # Enable progress notifications via messaging platforms\n  use_memory: false               # Enable cross-session knowledge persistence\n  use_web_fetch: false            # Enable live web search during literature review\n\n# === LLM ===\nllm:\n  provider: \"openai-compatible\"   # LLM provider: \"openai-compatible\" | \"openai\" | \"deepseek\" | \"acp\"\n  api_key: \"sk-your-key\"          # ⚠️ **IMPORTANT** API key (or use api_key_env to read from environment)\n  api_key_env: \"RESEARCHCLAW_API_KEY\"  # Environment variable name for API key (fallback)\n  primary_model: \"gpt-5.4\"        # ⚠️ **IMPORTANT** Main model for research, analysis, and writing\n  coding_model: \"gpt-5.4\"         # ⚠️ **IMPORTANT** Model for code generation (S11)\n  image_model: \"gemini-3-pro-image-preview\"  # ⚠️ **IMPORTANT** Model for figure generation in paper\n  fallback_models:                # Fallback model chain — used when primary model fails\n    - \"qwen3.5-plus\"\n    - \"qwen-plus\"\n\n# === Security ===\nsecurity:\n  hitl_required_stages: []        # Stage numbers requiring human approval (e.g. [5, 9, 20])\n\n# === Experiment ===\nexperiment:\n  mode: \"sandbox\"                 # Execution mode: \"sandbox\" (local Python) | \"docker\" | \"simulated\"\n  time_budget_sec: 2400           # ⚠️ **IMPORTANT** Max wall-clock time per experiment run (seconds)\n  max_iterations: 3               # Number of iterative refinement cycles in S15 (Edit-Run-Eval loop)\n  metric_key: \"primary_metric\"    # Name of the primary evaluation metric\n  metric_direction: \"minimize\"    # Optimization direction: \"minimize\" | \"maximize\"\n  datasets_dir: \"\"                # ⚠️ **IMPORTANT** Absolute path to datasets directory\n  checkpoints_dir: \"\"             # ⚠️ **IMPORTANT** Absolute path to model weights directory\n  codebases_dir: \"\"               # Absolute path to reference codebases directory\n  shared_results_dir: \"\"          # Directory for cross-project shared results\n  paper_length: \"long\"            # Paper length: \"short\" (~4 pages) | \"long\" (~8 pages)\n  sandbox:\n    python_path: \"\u002Fpath\u002Fto\u002Fpython3\"  # ⚠️ **IMPORTANT** Python interpreter for running experiments\n  sanity_check_max_iterations: 100   # Max fix attempts in S12 code testing\n\n# === Prompts ===\nprompts:\n  custom_file: \"\"                 # Path to custom prompts YAML file (empty = use defaults)\n```\n\n\u003C\u002Fdetails>\n\n\u003C!-- ---\n\n## Key Features\n\n\n- **Multi-Agent Discussion** | Multiple agents with different LLMs debate and reach consensus, avoiding homogeneous outputs. |\n- **Beast Mode Code Generation** | Complex experiments auto-routed to OpenHands for multi-file project generation. |\n- **Dynamic GPU Allocation** | Automatically detects free GPUs based on utilization. No manual `CUDA_VISIBLE_DEVICES`. |\n- **Checkpoint & Resume** | Auto-saves progress after each stage. Resume from any checkpoint after restart. |\n- **Manual Intervention** | Auto-pauses on code test failures. Yellow ⚠ indicator on UI with detailed error info. |\n- **Knowledge Loop** | Experiment results and insights feed back into the knowledge base for future projects. |\n- **Real-time Monitoring** | Web UI with agent status, GPU metrics, task queues, and event logs. |\n- **Paper with Figures** | Auto-generates experiment charts, renders concept figures, and injects them into the paper. | -->\n\n---\n\n## 🙏 Acknowledgement\n\nWe learned and reused code from the following projects: [AutoResearchClaw](https:\u002F\u002Fgithub.com\u002Faiming-lab\u002FAutoResearchClaw), [AutoResearch](https:\u002F\u002Fgithub.com\u002Fkarpathy\u002Fautoresearch), [claw-code](https:\u002F\u002Fgithub.com\u002Fultraworkers\u002Fclaw-code).\n\nWe thank the authors for their contributions to the community!\n\n## 📄 License\n\nMIT — see [LICENSE](LICENSE) for details.\n\n## 📌 Citation\n\nIf you find Claw AI Lab useful, please cite:\n\n```bibtex\n@misc{wu2026clawailab,\n  author       = {Wu, Fan and Chen, Cheng and Tan, Zhenshan and Zhang, Taiyu and Xu, Xinzhen and Qian, Yanyu and\n                  Gao, Dingcheng and Zhu, Lanyun and Zhu, Qi and Tan, Yi and Ji, Deyi and \n                  Lin, Guosheng and Chen, Tianrun and Ye, Deheng and Liu, Fayao},\n  title        = {Claw AI Lab: An Autonomous Multi-Agent Research Team},\n  year         = {2026},\n  url          = {https:\u002F\u002Fgithub.com\u002FClaw-AI-Lab\u002FClaw-AI-Lab},\n  note         = {GitHub repository}\n}\n```\n\n---\n\n## 💬 Community\n\n\u003Cp align=\"center\">\n  \u003Cimg src=\"assets\u002FGroup.png\" height=\"420\" alt=\"WeChat Group 1 QR Code\"\u002F>\n  \u003Cimg src=\"assets\u002FGroup2_new.png\" height=\"420\" alt=\"WeChat Group 2 QR Code\"\u002F>\n\u003C\u002Fp>\n","Claw AI Lab 是一个实验室原生的多智能体研究平台，旨在通过交互式和可扩展的人工智能驱动科学研究。其核心功能包括从单一指令创建完整的AI研究实验室，支持自定义角色、研究方向及协作工作流，并通过FIFO调度框架并行管理多个代理与项目，以最大化计算资源利用率。该平台还强调了人机协作的重要性，允许用户在必要时介入、提供反馈或引入新思路，进而迭代优化研究流程。此外，简洁直观的用户界面使得自动化研究过程更加可控且易于理解。适用于需要高效组织多智能体进行复杂科研任务的场景，如跨学科研究、大规模数据分析等。",2,"2026-06-11 03:49:39","high_star"]