Ryan Dahl Unveils celld – A Self‑Hosted Distributed Durable Objects Platform

Ryan Dahl, the founder of Node.js, released celld, a self‑hosted distributed implementation of the Cloudflare Workers and Durable Objects JavaScript API. The project aims to provide the same state‑management and operational capabilities of commercial serverless platforms without vendor lock‑in.

celld reproduces the Durable Objects model: each object has a unique global ID, its own SQLite replica, and runs inside a lightweight V8‑based VM. Storage is backed by Amazon S3 (or compatible) buckets, and the runtime uses the Rust Tokio async runtime. The codebase, written in Rust and JavaScript, is licensed under Apache 2 and can execute JavaScript/TypeScript directly; via WebAssembly it could also run Rust, C/C++, Go, or Zig with minimal V8 modifications.

Dahl claims celld’s operating costs are orders of magnitude lower than Cloudflare’s offering. He estimates that 100 persistent objects cost $415 per month on Cloudflare, whereas the same workload on celld—deployed on an 8 GB DigitalOcean Droplet with S3 storage—costs roughly $49 per month. Cloudflare later countered that if objects remain idle (sleep mode), the monthly cost drops to $20.65.

The platform provides a stateful serverless environment where data caching and computation coexist. It supports the WebSocket API, allowing persistent objects to maintain multiple simultaneous connections, and its sleep mode prevents charges when objects are unused. These characteristics make celld well‑suited for real‑time collaborative applications, multiplayer games, and AI agents. Cloudflare itself uses Durable Objects in its serverless SQL service and AI gateway.

Community response has been positive; a Hacker News reader praised the ability to run persistent objects outside a single provider, calling the abstraction valuable. Similar efforts are emerging, such as Neon Functions, which also colocate data and compute for long‑running workloads.

Although celld is open source, its GitHub repository explicitly bans AI‑generated contributions, stating that such submissions often lack context and increase maintenance effort. Human contributions are still welcomed, provided contributors understand the functionality before submitting code.