PilotDeck: An Open Source Agent OS with Transparent WorkSpaces

You've probably seen a dozen "agent frameworks" pop up this year. Most of them feel like black boxes. You send a prompt in, something happens, and an answer comes out. If you're building anything serious on top of LLMs, that opacity becomes a problem fast.

PilotDeck is different. It's an open source "agent OS" built around the idea of white box, traceable WorkSpaces. Think of it as a structured runtime for agents where you can actually see what's happening under the hood.

What It Does

PilotDeck gives you a platform to run, manage, and debug autonomous agents. Instead of a single agent doing everything in one shot, it uses the concept of WorkSpaces — isolated environments where agents operate with full observability.

You get:

• A runtime that manages agent lifecycles
• Built-in tooling for introspection and debugging
• A way to compose multiple agents or steps into workflows
• Full traceability of every decision an agent makes

It's essentially an operating system for your agents, not just a library.

Why It's Cool

Three things stand out here.

First, the white box design. Most agent frameworks give you a result and ask you to trust it. PilotDeck logs every step, every tool call, every piece of context the agent sees. When something goes wrong — and it will — you can trace exactly why.

Second, the WorkSpace abstraction. Agents operate inside sandboxed environments that you can inspect, pause, resume, or even snapshot. This makes debugging feel more like debugging a container than wrestling with a black box.

Third, it's built for developers who need to ship. The project is from OpenBMB (the same team behind other practical open source ML tools). The code is clean, well documented, and designed to be extended. If you need to add a custom tool or hook into the agent lifecycle, it's straightforward.

How to Try It

You can grab it from the GitHub repo right now.

git clone https://github.com/OpenBMB/PilotDeck.git
cd PilotDeck
pip install -e .

Check the README for quickstart examples. There's a demo that shows you how to set up a simple agent with a tool and trace its execution. No cloud account needed.

If you prefer to poke around before installing, the repo has a clean structure. The examples/ folder is a good place to start.

Final Thoughts

PilotDeck feels like the kind of tool you reach for when you're tired of "it works on my machine" for agents. The visibility into what the agent is actually doing is a breath of fresh air in a space full of magic boxes.

If you're building anything with LLMs that needs to be reliable, debuggable, or auditable — this is worth a look. It's early days, but the foundation is solid.

Found this useful? Follow us @githubprojects for more open source discoveries.