Ryan Dahl, who created Node.js, has released celld, an open source version of Cloudflare's Durable Objects technology that lets developers run the same stateful serverless workloads on their own infrastructure. Dahl announced the project last week, describing it as "a self-hosted, distributed Durable Objects and Workers implementation." The software is compatible with Cloudflare's JavaScript APIs but promises much lower operating costs, according to Dahl's team.
Dahl estimates that running 100 resident cells on celld would cost about $49 per month when built on a DigitalOcean S3-compatible bucket with an 8 GB droplet, compared to $415 monthly for the same workload on Cloudflare. Cloudflare disputed those figures, telling The Register that $415 would only apply if all objects were continuously active—if allowed to hibernate, the cost would drop to just $20.65 per month. The celld project swaps out Cloudflare's backend for Amazon S3 or equivalent storage and uses the Tokio Rust asynchronous runtime, while each celld object still gets its own SQLite copy. Written in Rust and JavaScript, the software is available under an Apache 2 license and can process JavaScript and TypeScript code, with theoretical support for Rust, C/C++, Go, or Zig through WebAssembly.
Dahl and his team call their work a "love letter" to the Durable Objects architecture, which they describe as "one of the best primitives distributed systems has been handed in years." The original technology, first devised by Kenton Varda and Cloudflare, has been in use since 2020 for building low-latency, highly distributed web applications. According to the celld creators, "a primitive this good deserves to run anywhere." Dahl promises the open source backend will be "orders of magnitude cheaper at scale" than Durable Objects, with further savings expected as workloads grow.
Durable Objects represents a shift from traditional serverless platforms like AWS Lambda by co-locating data with compute and using single-threaded execution to eliminate complex concurrency issues. Each object is a single-threaded entity with a unique global ID and its own storage, where user data lives in its own SQLite copy, running on Cloudflare's serverless Workers runtime. The model is a stateful serverless execution environment—essentially a data cache that can also do computation. Using the WebSocket API, a Durable Object can connect many simultaneous users at once in a live environment, and WebSockets' Hibernate mode can put the object to sleep so cloud bills don't pile up when nobody's using the app. This makes the technology well-suited for real-time collaborative applications like multiplayer games, team productivity tools, and AI agents—Cloudflare itself uses Durable Objects for its serverless SQL service and AI Gateway.
The concept appears to be gaining traction on its own merits beyond cost considerations. One Hacker News reader expressed enthusiasm about "support for running durable objects outside of one provider," noting the durable object is a valuable abstraction. Other companies are pursuing similar approaches to moving data closer to computation—Postgres service provider Neon recently introduced Neon Functions, which can also co-locate data and compute for long-running workloads. The celld project welcomes human contributions, though its GitHub page explicitly bans AI-generated code, stating that "coding agents make it too easy to send a large, low-context change that costs maintainers more time than it saves." Contributors should understand what their code does before submitting it. The arrival of an open source alternative could reshape how developers think about infrastructure lock-in, forcing cloud providers to compete not just on features but on the portability of their core primitives.