# PilotDeck Open-source agent operating system with WorkSpace isolation, white-box memory, smart routing, and always-on execution. ## Agent Decision Summary - Risk level: elevated - Source confidence: high - Recommended workflows: Coding agent workflow, Reusable skill workflow - Permission surface: memory, messages, external services - Agent JSON: https://www.openagent.bot/agents/pilotdeck.agent.json ## Summary PilotDeck is an open-source agent OS from Tsinghua THUNLP, ModelBest, OpenBMB, and AI9Stars, built around WorkSpace-level isolation for files, memory, and skills across projects. ## Guide PilotDeck is an open-source agent operating system designed around the concept of WorkSpace. Developed by Tsinghua THUNLP, ModelBest, OpenBMB, and AI9Stars, it targets general-purpose, multi-task productivity scenarios where agents need project isolation, traceable memory, cost-efficient routing, and unattended execution. ### What it is PilotDeck is an open agent OS to evaluate by project isolation and memory architecture: how it separates work into WorkSpaces, whether memory is white-box and editable, how it routes tasks to the right model tier, and whether it can run background discovery and cron jobs. ### Why it matters PilotDeck matters because it addresses the gaps in current agent platforms: memory that is opaque and uneditable, token costs that do not differentiate between trivial and complex tasks, and agents that stop working when you walk away. Its WorkSpace-first design and three pillar capabilities (white-box memory, smart routing, always-on) are practical explorations of these open problems. ### How it works Start with one WorkSpace in PilotDeck. Inspect how memory is captured and displayed, configure model providers with multiple tiers, enable TokenSaver routing, and test Always On discovery in an idle project. Expand from sandbox projects into production automation after verifying isolation and cost controls. ## Use Cases - Multi-project agent operations: Run parallel agent workflows where each project has independent files, memory, and skills without cross-contamination. - Cost-optimized agent deployment: Let TokenSaver automatically route simple tasks to cheaper models and complex tasks to flagship models, cutting bills by up to 70%. - Unattended background work: Enable Always On mode so the agent discovers code issues, runs cron jobs, and drops reports on disk while you are away. - Auditable agent memory: Use the white-box memory editor to view, edit, pin, and roll back exactly what the agent remembers per WorkSpace. ## Alternatives - Use OpenClaw or browser-use for browser automation vs browser agents: PilotDeck is a general-purpose agent OS. For pure browser or web workflow automation, OpenClaw or browser-use may be more specialized. - Use Claude Code or OpenHands for coding workflows vs coding agents: PilotDeck covers coding but is broader. Claude Code and OpenHands focus specifically on repository-level software engineering. ### Getting Started - One-line install: https://github.com/OpenBMB/PilotDeck - Official website: https://pilotdeck.openbmb.cn - Documentation: https://pilotdeck.openbmb.cn/pilotdeck.github.io/docs/en/introduction ### FAQ - What should I check before using PilotDeck? - Start with one WorkSpace in PilotDeck. Inspect how memory is captured and displayed, configure model providers with multiple tiers, enable TokenSaver routing, and test Always On discovery in an idle project. - Is PilotDeck open source? - PilotDeck is listed with AGPL-3.0 based on the official source links in this profile. Re-check the repository, license file, or docs before production use. - Who should evaluate PilotDeck? - PilotDeck is most worth evaluating for teams running multiple parallel agent projects that need clean isolation, traceable memory, and cost-efficient model routing. - What makes PilotDeck different from other agent platforms? - PilotDeck is WorkSpace-first: each project has isolated files, memory, and skills. It also provides white-box editable memory, automatic model tiering (TokenSaver), and always-on background execution — capabilities most agent platforms lack. ## What It Does PilotDeck is an open agent OS to evaluate by project isolation and memory architecture: how it separates work into WorkSpaces, whether memory is white-box and editable, how it routes tasks to the right model tier, and whether it can run background discovery and cron jobs. ## How To Evaluate Start with one WorkSpace in PilotDeck. Inspect how memory is captured and displayed, configure model providers with multiple tiers, enable TokenSaver routing, and test Always On discovery in an idle project. Expand from sandbox projects into production automation after verifying isolation and cost controls. ## Why It Matters PilotDeck matters because it addresses the open problems of production agent work: memory traceability, per-task cost tracking, automatic model tiering, and background execution when you walk away from the keyboard. ## Best For - Teams running multiple parallel agent projects that need clean isolation - Builders who want white-box, editable, rollback-capable agent memory - Users looking to cut token costs through intelligent model routing ## Not For - Users who only want a single-chat chatbot with no project structure - Teams unwilling to review model provider credentials and routing configuration before usage ## What It Actually Does - WorkSpace-level isolation: Every project gets its own file system, memory store, and skill set. Parallel work does not interfere, retrieval is bounded, and skills accrete naturally per project. - Why it matters: Without isolation, multiple tasks pollute a shared context; with PilotDeck, projects run in parallel with clean boundaries. - Traceable white-box memory: Memory generation, extraction, storage, and retrieval are visible end-to-end. Users can edit or delete entries, pin critical decisions, and one-click rollback via Dream Mode. - Why it matters: When AI mis-remembers, you can pinpoint the offending entry and fix it directly instead of starting a new chat. - Smart routing for cost optimization: TokenSaver automatically classifies task complexity and routes simple tasks to lighter models, cutting token spend by up to 70% without sacrificing quality. - Why it matters: Running agents in the background becomes economically viable when you are not burning flagship models on trivial calls. - Always-on background execution: After you sign off, the agent keeps discovering candidate tasks, running monitors, and landing deliverables as files with a summary report waiting for you. - Why it matters: Agent work should not stop when you close the laptop. ## Typical Use Cases - Multi-project agent workflows: Use PilotDeck when running agents across several projects simultaneously, with each needing independent files, memory, and skills. - Cost-sensitive agent deployment: Use TokenSaver routing to automatically match task difficulty to the right model tier, dramatically cutting token bills. - Unattended automation: Deploy Always On mode so the agent discovers issues, runs scheduled cron tasks, and produces reports while you are away. - Auditable agent memory: Use white-box memory when you need to inspect, edit, or roll back exactly what the agent remembers per project. ## How It Compares - PilotDeck is a full agent OS, not just a coding agent vs coding agents: Claude Code and OpenHands focus on repository-level coding. PilotDeck is a general-purpose agent OS with WorkSpace isolation, routing, memory, and always-on execution. - PilotDeck isolates by project, not by session vs standard chat agents: Standard agents mix all context in one pool. PilotDeck gives each WorkSpace its own files, memory, and skills so projects never bleed into each other. ## Fit Matrix - Coding agent workflow: strong. PilotDeck has multiple signals for coding agent workflow, including matching tags, capabilities, category, or positioning. Required check: Run a small repository change and inspect the diff, tests, and rollback path. - Reusable skill workflow: strong. PilotDeck has multiple signals for reusable skill workflow, including matching tags, capabilities, category, or positioning. Required check: Run one skill end to end and check whether it produces evidence or structured output. - Connector or protocol layer: partial. PilotDeck has at least one signal for connector or protocol layer, but should be checked against a real task before adoption. Required check: Connect one low-risk service, then inspect schemas, auth scope, errors, and logs. - Local or private AI stack: partial. PilotDeck has at least one signal for local or private ai stack, but should be checked against a real task before adoption. Required check: Verify hardware requirements, data path, storage, and whether all calls stay in your environment. - Memory or RAG workflow: partial. PilotDeck has at least one signal for memory or rag workflow, but should be checked against a real task before adoption. Required check: Create, update, retrieve, correct, and delete memory or retrieval objects with real data. - Browser automation: weak. PilotDeck is not primarily positioned for browser automation in the current metadata. Required check: Run one non-sensitive website task and inspect clicks, waits, retries, and changed URLs. ## Evidence - verified: PilotDeck is listed as open source. Source: License metadata: AGPL-3.0 - verified: PilotDeck has a recorded GitHub repository: OpenBMB/PilotDeck. Source: Resource facts and GitHub source link. - inferred: PilotDeck supports these recorded deployment modes: self hosted, cloud. Source: OpenAgent decision signal metadata. - inferred: PilotDeck is tagged with memory, mcp, workflow orchestration capabilities. Source: OpenAgent capability taxonomy. ## Missing Checks - Repository freshness has not been recorded. ## Next Actions - Inspect repository: https://github.com/OpenBMB/PilotDeck - Open Homepage: https://pilotdeck.openbmb.cn - Read setup docs: https://pilotdeck.openbmb.cn/pilotdeck.github.io/docs/en/introduction - One-line install (macOS / Linux): curl -fsSL https://raw.githubusercontent.com/OpenBMB/PilotDeck/main/install.sh | bash - Start the server: pilotdeck ## Command Line ### One-line install (macOS / Linux) Auto-installs Node.js 22, clones the repo, installs dependencies, and builds the frontend. ```bash curl -fsSL https://raw.githubusercontent.com/OpenBMB/PilotDeck/main/install.sh | bash ``` ### Start the server Starts the PilotDeck server at http://localhost:3001 ```bash pilotdeck ``` ### Clone from source Manual source-based install for developers. ```bash git clone https://github.com/OpenBMB/PilotDeck.git && cd PilotDeck && npm install && cd ui && npm install && cd .. ``` ## Facts - Category: agents - Resource type: agent - Open source: yes - License: AGPL-3.0 - Last verified: 2026-06-03 - GitHub repo: OpenBMB/PilotDeck - GitHub stars: 2873 ## Capabilities - memory - mcp - workflow-orchestration ## Structured Use Case Tags - self-hosted-ai - developer-workflow ## Getting Started - Review the GitHub repository: https://github.com/OpenBMB/PilotDeck - Visit the official website: https://pilotdeck.openbmb.cn - Read the docs: https://pilotdeck.openbmb.cn/pilotdeck.github.io/docs/en/introduction ## Links - GitHub: https://github.com/OpenBMB/PilotDeck - Homepage: https://pilotdeck.openbmb.cn - Docs: https://pilotdeck.openbmb.cn/pilotdeck.github.io/docs/en/introduction ## Structured Outputs - JSON: https://www.openagent.bot/agents/pilotdeck.json - Markdown: https://www.openagent.bot/agents/pilotdeck.md - Agent JSON: https://www.openagent.bot/agents/pilotdeck.agent.json - Canonical: https://www.openagent.bot/agents/pilotdeck