Skip to main content

Cloudflare Workers-style infrastructure without the cloud

Kagi Small Github
15 lis 2026, 21:57Tech
Oryginał
2 min read

A new Rust-based platform called open-compute allows developers to run Cloudflare Workers-compatible applications on their own hardware with a single binary.

Open-compute is a self-hosted platform that brings Cloudflare Workers compatibility to personal hardware through a single Rust binary. The project aims to eliminate vendor lock-in and complex infrastructure dependencies while maintaining the familiar Workers programming model developers already know.

Unlike Cloudflare's global edge network, open-compute runs on a single machine owned by the user. This trade-off provides strong local consistency without the need for Anycast or cross-region replication. The platform supports ten Cloudflare products including Workers, KV, D1, R2, Durable Objects, Queues, Vectorize, AI Search, and Observability.

Simplified Architecture

The platform's architecture is notably simplified compared to traditional solutions. It requires no Kubernetes, Redis, service mesh, or distributed control plane. Instead, open-compute delivers everything in one daemon with SQLite handling platform metadata and local storage managing object bytes by default.

Performance is prioritized through the use of workerd, a Cloudflare runtime fork. Worker code runs on a pinned, checksum-verified version, with isolated instances starting in milliseconds rather than requiring separate processes or containers for each request.

One binary. One shared daemon. One object authority per instance. Local filesystem is the default; S3-compatible storage is optional.

elliothux, developer

Compatibility and Setup

Open-compute maintains API compatibility with Cloudflare Workers, supporting module workers and familiar bindings. The test suite tracks 2,256 stable API members and overloads, comparing behavior with real Cloudflare where possible. Incompatible features are rejected rather than approximated.

Setup is straightforward, with both manual and AI-assisted options. The manual process involves downloading the binary, creating a default instance, and starting the service. For production, the platform maintains a single release executable with explicit per-instance configuration and embedded runtime payloads.

Security and Extensions

Security is enforced through exclusive locks for each instance and internal tokens that never appear in logs or environment variables. Tenant outbound connections can be filtered through host firewall controls. The platform also supports native extensions for Workers requiring local hardware or proprietary libraries.

The project is licensed under Apache-2.0, with the packaged workerd fork retaining its original Cloudflare licensing. Development continues with sponsorship from Lynx AI, and security vulnerabilities can be reported privately through the project's SECURITY.md file.

Z.ai
Article written by AI
Z.ai GLM-4.5-FlashFree
Free AI articles150 of 150
No limit after signing in with Telegram