PilotDeck - Tsinghua University's joint open-source Agent operating system with Facetower Intelligence - AiBoss

What is PilotDeck?

PilotDeck is an open-source intelligent agent operating system jointly developed by Tsinghua University's THUNLP Lab, Facewall Intelligence, OpenBMB, and AI9stars. PilotDeck uses the WorkSpace design paradigm as its core, establishing independent file systems, memory storage, and skill sets for each project to achieve multi-task parallelism and physical isolation. The system supports white-box memory, intelligent routing for dynamic model allocation, and 24/7 always-on persistent tasks, achieving a cost reduction of over 70% in real-world testing. It also supports edge-cloud collaboration and automatic deployment of edge models, evolving AI from a conversational toy into a true productivity tool.

PilotDeck's main functions

WorkSpaceEstablish an independent living space for each project, including a dedicated file system, memory, and skills, to achieve project-level physical isolation.
Memory white boxThe entire memory chain is visible, controllable, and traceable. Users can view, accurately locate, and correct or delete erroneous memories at any time with one click.
Smart RouterAutomatically identifies task difficulty and dynamically allocates models; simple tasks call lightweight client-side models, while complex tasks call top-level cloud models.
Always-on missionsThe agent proactively identifies incomplete tasks and autonomously advances them without requiring continuous manual triggering, enabling it to work online 24/7.
End-to-cloud collaborationIt supports calling edge-side models as sub-Agents to execute specific tasks, and can automatically deploy edge-side models (such as VoxCPM) to complete tasks such as speech generation.

PilotDeck's technical principles

Three-layer architecture of the work cabinEach WorkSpace consists of a dedicated file system, dedicated memory storage, and a dedicated skillset, achieving project-level resource boundary isolation.
Memory white box mechanismIt adopts full-link memory storage and visual traceability, supports task-based Dream automatic sorting and optimization, and has one-click rollback capability to prevent memory pollution.
Intelligent routing algorithmBased on the automatic classification of task complexity, the strong and weak models are dynamically scheduled through the orchestration of the main agent and sub-agents, thus optimizing the balance between cost and effect.
Resident mission engine: Give the Agent initiative, enabling it to autonomously scan project status, discover pending tasks, perform operations, and generate output documents.
Cross-project knowledge transferIt supports the evolution and accumulation of skills as tasks progress, enabling controlled knowledge transfer and reuse across different Workspaces.

How to use PilotDeck

One-click installationTo install, run the official curl command in the macOS or Linux terminal.
Start service:implement pilotdeck Command starts, browser access http://localhost:3001 Enter the console.
Create a work podCreate a new WorkSpace in the web interface and set the project name and file boundaries.
Configure skillsInstall the skill packs required for the project with one click; the Agent will automatically accumulate exclusive capabilities as the task progresses.
Issue instructionsEnter the natural language task in the dialog box, and the agent will automatically plan, execute, and continuously advance it. Users can view the memory and progress at any time.

PilotDeck's project address

Project official website: https://pilotdeck.openbmb.cn/pilotdeck.github.io/
GitHub repositoryhttps://github.com/OpenBMB/PilotDeck

PilotDeck's Competitive Product Comparison

Dimension	Pilot Deck	OpenClaw
Core Architecture	WorkSpace = Project-level OS	Gateway = Message Channel Hub
Permission Model	Free operation within the work compartment, with physical isolation between compartments.	Full system access, with both risks and capabilities.
Memory strategies	White box memory + automatic cleanup + one-click rollback	Black-box vector memory, long-term persistence
initiative	The agent proactively identifies and advances tasks.	Heartbeat timer trigger + message-driven
cost	Built-in smart router automatically saves money	Relying on users to manually select models
End side	Native automatic deployment of end-side models	Ollam needs to be configured manually.
Safety	The boundaries are clear, and the impact of misoperation is controllable.	Excessive permissions require manual security measures.

PilotDeck application scenarios

Game developmentGenerate an interactive 3D open-world game (like Zelda) with rich elements and physical interactions using client-side models within a day.
Multilingual podcast generationIt integrates with the VoxCPM voice model to generate multilingual podcast content supporting 30 languages with a single click.
Mobile application developmentRapidly develop prototypes of lightweight applications such as mobile gyroscope-based games.
AI model trainingIt automates the model training process and supports continuous background operation and progress tracking for long-cycle tasks.
Long-term content creationUsed for complex document projects that require continuous iteration and memory accumulation, such as writing industry white papers and creating scripts.