Building a decentralized OS from scratch is a massive
What is Ecdeos actually doing?
The project uses a Tauri Rust/WASM browser engine combined with a Rust backend. Instead of a traditional central server, it treats everything as an "entity." By using content-addressed data and a capability-based security model, it aims to scale from a single user managing personal devices to massive federated networks.
It's drawing inspiration from a heavy-hitter list of technologies:
- Plan9 and Erlang/BEAM for system philosophy.
- IPFS and ATProto for decentralized data and identity.
- Nostr for the networking layer.
The Technical Stack and Deployment
The architecture is split between web-based deployments and native backends. If you're looking at the implementation details, here is how the environment is currently structured:
- Web Environment: Utilizes WebSockets and WebRTC for P2P chat, file transfers, and entity movement. It supports "coral reef" mode, meaning it can run without a live peer by pulling signed sites from a CDN-hosted registry and caching them locally.
- Tauri Native Backend: This is where the heavy lifting happens, adding support for SQLite storage, TCP, and WebSockets.
- Application Layer: The web deployment ships with over 30 "Entity-Apps" (games, music, tools) running in isolated iframe sandboxes. These are currently JS-based, but they also feature Entity-Native compute apps running as Rust WASM peers.
One interesting detail for the devs here is the compute language. It’s a transferable system language that runs on a cross-platform generic system host interface managed by the engine. They actually use a separate Go-based workbench project to compile and test this compute code in parallel to the main implementation.
Real-world implications for AI workflows
The developers were surprisingly blunt about one thing: almost all the code and content is AI-generated. While they joked about it being "the best slop available," from a prompt engineering and LLM agent perspective, it’s a fascinating case study. They are essentially using AI to build the very infrastructure that might one day host autonomous agents in a decentralized environment.
If we eventually move toward a world where LLM agents need to own their own data and compute without relying on a centralized AWS or Google instance, a protocol like this—where the "OS" is distributed across the peers themselves—might be the only way to maintain sovereignty.
It’s still a research prototype and definitely needs stress testing, but the design seems robust enough to handle the transition from local storage to a full P2P network.
https://ecdeos.org
https://entitycoreprotocol.org