AI NewsInfrastructureAnnouncement

open-compute runs the Cloudflare Workers platform on your own hardware

open-compute packages the Cloudflare Workers runtime, KV, R2, D1, Durable Objects, Queues and Workflows into one self-hosted binary, and made its first release on 5 September with 425 stars in under two weeks.

AI News

Editorial2 min read

LinkedInX
The GitHub card for the elliothux/open-compute repository

Image: GitHub

Why it mattersA Workers app can now run air-gapped or on your own servers without a rewrite, which changes whether choosing the platform ties you to one vendor.

open-compute is a single binary that runs Cloudflare Workers and most of the storage products around it on your own machine. The project published its first tagged release, v0.1.0, on 5 September 2026. The GitHub API showed 425 stars and an Apache-2.0 licence when it was checked the same day, against a repository created on 23 August, so that interest came in under two weeks.

It is built on workerd, the V8 runtime Cloudflare itself open-sourced. The part that did not exist before is everything above the runtime: the control plane, the scheduler and the bindings, which Cloudflare has never released as something you can run yourself.

What the coverage claim actually says

The README claims 2,097 stable API members implemented. It lists per-surface completion rather than one headline number, which is the useful form. Workers, KV, R2, D1, Durable Objects, Queues, Cron, Workflows and Cache are each marked 100 percent. The local Cloudflare v4 API is at 90 percent, and Wrangler compatibility at 95 percent, pinned to version 4.127.1.

Three surfaces are marked partial at 80 percent: Vectorize, Markdown Conversion and AI Search. Workers AI, Containers, Hyperdrive and Analytics Engine are not implemented at all. Those are the project's own figures, and nobody independent has audited them, so treat the table as a statement of intent that is at least specific enough to check against your own wrangler config.

The limits are stated up front

This is one node. The README says so directly: no distributed cluster, no high availability, no global edge replication, and egress is local only. Anything you deploy to it reaches users from wherever that machine is, which removes the entire reason most teams chose Workers in the first place.

Performance is described as millisecond cold starts and MB-scale memory for idle isolates, with no measured figure attached to either phrase. The project reports 90 percent enforced line coverage in its acceptance runs, which is a claim about its own test suite rather than about compatibility. Building from source needs Rust 1.98, Bun 1.3 and Node 24, on macOS or Linux.

The value is in the decision it makes possible

Almost nobody is going to move production traffic off Cloudflare's edge onto a single self-hosted machine. The more useful point is that "we cannot run this anywhere else" stops being true. A team that rejected Workers because a customer requires an air-gapped install, or because the compliance answer to "what happens if this vendor changes terms" was nothing, now has a different answer. The same applies to CI: running the real bindings instead of a simulation makes an integration test mean something closer to what production does.

Version 0.1.0 on day one, with three surfaces incomplete and four missing, is early. The honest view is that this is worth testing your build against now and worth nothing as a fallback until somebody has actually run it under load. Both the compatibility table and the coverage figure come from the project itself, and the first useful independent signal will be somebody reporting which of their existing Workers failed to start.

Source

elliothux/open-compute, GitHub. Release v0.1.0 published 5 September 2026.

SourceGitHub

This item was written by an AI system from the linked source. Reveneau is responsible for what it publishes.

Share
LinkedInX
Start a project