{"id":"5318a823-48d3-4964-8675-02ace43a9b79","slug":"pixelagency","kind":"skill","name":"PixelAgency","summary":"๐ŸŽฎ 215ไฝAIๆ™บ่ƒฝไฝ“่™šๆ‹ŸๅŠžๅ…ฌๅฎค - ๅƒ็ด ้ฃŽๆธธๆˆๅŒ–Agent่ฐƒๅบฆ็ณป็ปŸ","owner":"feifeida57","repoUrl":"https://github.com/feifeida57/PixelAgency","license":"AGPL-3.0","licenseTier":"strong-copyleft","trust":"community","stars":1,"match":null,"fewStars":true,"format":"skill","createdAt":"2026-05-31T14:27:32.000Z","addedAt":"2026-10-21T21:17:07.169Z","logoUrl":"https://avatars.githubusercontent.com/u/281360036?v=4","linkOnly":false,"collection":null,"skillPath":null,"categories":["writing"],"tags":[],"runtime":"binary","local":true,"cloudCapable":false,"cloudCapability":"unknown","cloudReadyOnPlatform":false,"needsApiKey":false,"reviewed":false,"syncedAt":"2026-10-23T12:40:14.473Z","sample":false,"url":"https://conduits.dev/tools/pixelagency","language":"TypeScript","transports":["stdio"],"install":{},"requirements":{"env":[],"upstreamMayCharge":false},"tools":null,"readme":"

\n \"PilotDeck\"\n

\n\n

\n Task-oriented AI Agent productivity platform โ€” redefining operational boundaries and memory evolution, one WorkSpace at a time.\n

\n\n

\n \"Official\n \"Live\n \"License\"/\n \"MCP\n \"Stars\"/\n
\n \"Discord\"/\n  \n \"Feishu\"/\n  \n \"WeChat\"/\n
\n

\n\n

\n English | ็ฎ€ไฝ“ไธญๆ–‡\n
\n Website ยท Live Demo ยท Tutorial ยท Quick Start ยท Highlights ยท Use Cases ยท Community\n

\n\n---\n\n**News** ๐Ÿ”ฅ\n\n- **[2026.05.28]** PilotDeck is now open source! Visit our official website at [pilotdeck.openbmb.cn](https://pilotdeck.openbmb.cn). We welcome contributions, feedback, and stars from the community.\n\n---\n\n## ๐Ÿ’ก About PilotDeck\n\n**PilotDeck** is an open-source agent operating system designed around the concept of \"WorkSpace\". It is jointly developed and open-sourced by Tsinghua University [THUNLP](https://nlp.csai.tsinghua.edu.cn/), [ModelBest](https://modelbest.cn/), [OpenBMB](https://www.openbmb.cn/), and [AI9Stars](https://github.com/AI9Stars). Targeting general-purpose, multi-task scenarios, PilotDeck is built to be a true *productivity tool* for the Agent era.\n\nA wave of excellent AI Agent harnesses has emerged in recent years, each with its own focus: **Claude Code / Cursor / Trae Solo** brought model reasoning deep into the programming IDE; **Claude Cowork** introduced the notion of project-level isolation to desktop-side knowledge work; **WorkBuddy** connected agents to IM ecosystems such as WeCom and Feishu so AI is one message away.\n\nWhen we shift the lens from \"one-shot programming\" or \"immediate Q&A\" to **long-running, multi-project productivity work**, however, several questions remain open:\n\n- When many projects run in parallel, can memory be **white-box and traceable**? When the AI gets something wrong, can you pinpoint which memory entry caused it and edit it directly โ€” without starting a new chat from scratch?\n- Can token cost be **tracked per task**, so that running agents in the background actually becomes economically viable?\n- Can tasks of different difficulty **automatically be matched to different models**, instead of burning the flagship model on trivial calls?\n- When you step away from the keyboard, can the work keep moving? Can the agent **proactively discover what's worth doing, report progress, and land results as files on disk**?\n\nPilotDeck is an incremental exploration around exactly these questions. It uses the WorkSpace as the fundamental unit โ€” completely isolating files, memory and skills per project โ€” and pairs it with three pillar capabilities: **White-box Memory**, **Smart Routing** and **Always-on**. The entire system natively supports the [Model Context Protocol (MCP)](https://modelcontextprotocol.io/) and behaves consistently across front-ends (Web / CLI / IM).\n\n### โœจ Key Highlights\n\n\n\n\n\n\n\n\n\n\n
\n\n**WorkSpace-Level Isolation & Accretion**\n\nEvery project gets its own file system, memory store and skill set. Parallel work no longer interferes with itself, retrieval has a bounded scope, and skills accrete naturally as each task grows โ€” no more global context pollution.\n\n

\n \"WorkSpace\n

\n\n		\n\n**Traceable White-box Memory**\n\nMemory generation, extraction, storage and retrieval are visible end-to-end. When the AI mis-remembers, you can pinpoint and fix the offending entry. Built-in **Dream Mode** consolidates memory in idle windows, and supports one-click rollback.\n\n

\n \"White-box\n

\n\n
\n\n**Smart Routing & Cost Optimization**\n\nTask difficulty is auto-detected; complex calls go to flagship models (e.g. Claude 3.5 Sonnet / GPT-4o), simple ones drop to lighter models. Through on-device / cloud co-orchestration and precise matching, token spend shrinks dramatically without sacrificing quality.\n\n

\n \"Smart\n

\n\n		\n\n**Always-on Background Execution**\n\nPilotDeck breaks the \"you ask, it answers\" loop: after you sign off, the agent keeps discovering candidate tasks, running long-horizon monitors, and finally lands deliverables as local files with a summary report waiting for you.\n\n

\n \"Always-on\n

\n\n
\n\n### ๐Ÿ“Š Real-world Numbers\n\nThe three pillar capabilities have shown clear advantages in production-grade workflows:\n\n#### 1. Smart Routing โ€” ~70% cost savings on social-media workloads\n\nIn Xiaohongshu-style social-media operations, enabling Smart Routing automatically demotes simple polishing / layout tasks to a sub-agent (e.g. Sonnet 4.5) and only invokes Opus 4.5 at planning checkpoints:\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n
SetupModel configurationCostMultiplier
Smart Routing ONOpus 4.5 (main) + Sonnet 4.5 (sub)$2.831.1ร—
Smart Routing OFFAll Opus 4.5 (main + sub)$12.585.0ร—
MonolithicSingle Opus 4.5 long-react (estimated)$12.204.8ร—
\"\"/
\n\n#### 2. Smart Routing โ€” 1/6 the cost while beating frontier models on hard tasks\n\nThe research team benchmarked 7 complex tasks (multilingual podcast push, multi-source data reports, domain-specific literature review, codebase architecture docs, etc.). The \"strong main + light sub\" routing setup matches or beats the frontier single-model setup at a fraction of the cost:\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n
SettingScoreCost
MiniMax-M2.7 single-agent37.1$1.90
Claude Sonnet 4.6 single-agent69.1$18.36
Sonnet 4.6 (main) + MiniMax-M2.7 (sub)70.6$3.15
\"\"/
\n\n#### 3. White-box Memory โ€” layout & tone never bleed across projects\n\nIn black-box agents, mixing tasks in a shared context pool inevitably pollutes memory. PilotDeck's WorkSpace-scoped white-box memory addresses this end-to-end:\n\n\n\n\n \n \n \n\n\n\n\n \n \n \n\n\n \n \n \n\n\n \n \n \n\n\n \n \n \n\n\n \n \n \n\n\n
DimensionCurrent AI Agents (black-box)PilotDeck (white-box)
VisibilityYou can't see what the AI remembers, only what it outputsView every memory entry: what was stored, when, and which WorkSpace
ControlOnce written, memory can't be edited or removedEdit / delete entries, pin critical decisions so they don't drift
TraceabilityWhen it goes wrong, you can't find the root causeGeneration โ†’ extraction โ†’ storage โ†’ retrieval, all auditable
IsolationOne shared pool โ€” projects bleed into each otherScoped per WorkSpace; A's memory never reaches B
ReversibleAfter compression, the original is goneDream-mode supports one-click rollback to the prior state
\n\n---\n\n## ๐Ÿ–ฅ๏ธ UI & Demo\n\nPilotDeck ships an out-of-the-box Web UI with full WorkSpace management, white-box memory editing, and visualization of multi-agent collaboration.\n\n### Use Cases\n\n> All demos below are generated entirely by edge-side models via PilotDeck's Smart Routing โ€” no cloud-side frontier model required.\n\n#### Work Document Generation\n\n> *\"Survey the Chinese LLM application market and turn it into a formal HTML white paper.\"*\n\n\n\n\n\n\n\n\n\n\n
ProcessResult
\n\n#### Mini-Game Development\n\n> *\"Walk me through building an iOS AR mini-game Ball Finder in Vibe Coding mode.\"*\n\n\n\n\n\n\n\n\n\n\n
ProcessResult
\n\n#### AI Engineering Platform Development\n\n> *\"Build a low-code embedding fine-tuning platform from scratch.\"*\n\n\n\n\n\n\n\n\n\n\n
ProcessResult
\n\n#### Audio-Video Editing & Social Media Operations\n\n> *\"Push this English podcast to a global audience in Chinese / Japanese / French / Korean / Spanish / Arabic.\"*\n\n\n\n\n\n\n\n\n\n\n
ProcessResult (with audio)
\n\nhttps://github.com/user-attachments/assets/a7245467-ee3c-4939-a055-c56576ac56d1\n\n
\n\n---\n\n## ๐Ÿ“ฆ Installation & Quick Start\n\nWe provide a one-line installer for macOS / Linux, plus a source-based workflow for developers.\n\n### Option A: One-line install (recommended, macOS / Linux)\n\n```bash\ncurl -fsSL https://raw.githubusercontent.com/OpenBMB/PilotDeck/main/install.sh | bash\n```\n\nThe script auto-installs Node.js 22, clones the repo, installs dependencies, and builds the frontend. Once it finishes:\n\n```bash\npilotdeck # starts the server at http://localhost:3001\npilotdeck status # check runtime status\n```\n\n### Option B: From source (for developers)\n\n**1. Clone and install dependencies**\n\n> This repo uses [Git LFS](https://git-lfs.com/) for large media assets. Make sure `git lfs` is installed before cloning.\n> If you don't need the demo videos/GIFs, add `GIT_LFS_SKIP_SMUDGE=1` before `git clone` to skip downloading them.\n\n```bash\ngit clone https://github.com/OpenBMB/PilotDeck.git\ncd PilotDeck\n\nnpm install # root deps (Gateway runtime)\ncd ui && npm install # UI deps\ncd ..\n```\n\n**2. Configure a model provider**\n\nPilotDeck reads `~/.pilotdeck/pilotdeck.yaml`. You can create it manually, let the bootstrap script generate one, **or just open the Web UI and configure providers visually in the settings panel.**\nSupported protocols include OpenAI, Anthropic, DeepSeek, Qwen, Kimi, MiniMax and other OpenAI-compatible endpoints.\n\n```yaml\nschemaVersion: 1\nagent:\n model: deepseek/deepseek-v4-pro\nmodel:\n providers:\n deepseek:\n protocol: openai\n url: https://api.deepseek.com/v1\n apiKey: sk-your-api-key\n```\n\n**3. Start the services**\n\n```bash\ncd ui && npm run dev # dev mode (HMR), visit http://localhost:5173\n# or\ncd ui && npm run start # production mode, visit http://localhost:3001\n```\n\n### Option C: Docker Compose\n\nIf Docker is installed, you can start PilotDeck with:\n\n```bash\ndocker compose up -d\n```\n\n---\n\n## ๐Ÿ› ๏ธ Extension Protocol\n\nPilotDeck has an open plugin architecture with a strict boundary between the open-source core and plugin customization. Extending the system is a `plugin.json` away:\n\n- **MCP Servers** โ€” first-class integration with any Model Context Protocol server.\n- **Tools & Skills** โ€” register custom tools, or pull community skills via [ClawHub](https://www.npmjs.com/package/clawhub).\n- **Lifecycle Hooks** โ€” intercept `PreToolUse`, `UserPromptSubmit`, and other critical lifecycle events.\n- **Custom Memory** โ€” plug in your own memory store provider.\n\n---\n\n## ๐Ÿค Contributing\n\nThanks to everyone who has contributed code, feedback, and ideas. New contributors are warmly welcome โ€” let's build the next-gen agent OS together.\n\nWorkflow: **Fork โ†’ feature branch โ†’ PR**.\n\n---\n\n## ๐Ÿ’ฌ Community\n\n- For bugs and feature requests, please open a [GitHub Issue](https://github.com/OpenBMB/PilotDeck/issues).\n- Join our community channels:\n\n\n\n\n\n\n\n\n\n\n\n\n
WeChat CommunityFeishu CommunityDiscord Community
\"WeChat\"Feishu\"Discord
\n\n---\n\n## ๐Ÿ™ Acknowledgements\n\nWe thank Agent OS pioneers such as OpenClaw, Claude Code, Codex, Cursor, and Hermes for their explorations that helped shape this field.\n\nPilotDeck builds upon the following outstanding open-source projects:\n\n- [ClawXRouter](https://github.com/OpenBMB/ClawXRouter) โ€” Intelligent model routing\n- [ClawXMemory](https://github.com/OpenBMB/ClawXMemory) โ€” Agent memory system\n- [Claude Code UI](https://github.com/siteboon/claudecodeui) โ€” Web UI reference\n- [Claude Code Router](https://github.com/musistudio/claude-code-router) โ€” Model routing reference\n- [UltraRAG](https://github.com/OpenBMB/UltraRAG) โ€” RAG framework\n- [Anthropic Skills](https://github.com/anthropics/skills) โ€” Agent skill framework and built-in skills (skill-creator)\n- [Vercel Labs Skills](https://github.com/vercel-labs/skills) โ€” find-skills skill\n- [MiniMax-AI Skills](https://github.com/MiniMax-AI/skills) โ€” minimax-pdf skill\n- [frontend-slides](https://github.com/zarazhangrui/frontend-slides) โ€” Create beautiful slides on the web using a coding agent's frontend skills\n- [Karpathy Guidelines](https://x.com/karpathy/status/2015883857489522876) โ€” LLM coding behavioral guidelines\n- [Vite](https://github.com/vitejs/vite) โ€” Frontend build tool\n- [React](https://github.com/facebook/react) โ€” UI framework\n- [Tailwind CSS](https://github.com/tailwindlabs/tailwindcss) โ€” Utility-first CSS framework\n- [shadcn/ui](https://github.com/shadcn-ui/ui) โ€” Accessible component primitives for React\n\n---\n\n## ๐Ÿข Joint Development\n\nPilotDeck is jointly developed by Tsinghua University [THUNLP](https://nlp.csai.tsinghua.edu.cn/), [ModelBest](https://modelbest.cn/), [OpenBMB](https://www.openbmb.cn/) and [AI9Stars](https://github.com/AI9Stars).\n\n---\n\n## โญ Support Us\n\nIf PilotDeck has been helpful in your work or research, please consider giving us a Star on GitHub!\n\n---\n\n## ๐Ÿ“ Citation\n\n```bibtex\n@misc{pilotdeck2026,\n author = {PilotDeck Team},\n title = {PilotDeck: A WorkSpace-Centric Open-Source Agent Operating System},\n howpublished = {\\url{https://github.com/OpenBMB/PilotDeck}},\n year = {2026},\n note = {Accessed: 2026-05-29}\n}\n```\n\n## ๐Ÿ“„ License\n\nThis project is licensed under the [GNU Affero General Public License v3.0](LICENSE).\n"}