Introducing Forge: the open source pipeline for generating SDKs, CLIs, docs, and more

Cloudflare open‑sourced Forge, a pluggable pipeline that generates SDKs, CLIs, documentation and other artifacts from API specs. The blog post shows it running in CI to lint changes and produce preview builds for Cloudflare’s 3,500‑operation API, supporting Rust, Go, TypeScript and Python. Developers can deploy it free to automate multi‑language client generation.

Cover image for Introducing Forge: the open source pipeline for generating SDKs, CLIs, docs, and more

Cloudflare announced the open‑source release of Forge, a pluggable pipeline that can generate SDKs, CLIs, documentation and other artifacts directly from API specifications. The system runs in continuous integration, linting each change and producing preview builds that highlight only the modifications. In its current form it supports Rust, Go, TypeScript and Python and is already used to generate the cf CLI for an API that contains more than 3,500 operations. Forge is designed to cut coordination overhead between product teams by allowing every team to test generated client code before merging API changes. By treating the generation step as a first‑class part of the CI workflow, it reduces the risk of breaking the pipeline and makes it possible to extend generation to custom targets such as Cap’n Web, TanStack Query bindings or schema libraries. Cloudflare plans to use the tool for its own API documentation, SDKs and other surfaces over the next few months, and it can be deployed free of charge by any developer. The project is still early in its lifecycle, and many of the longer‑term claims remain unproven. Cloudflare’s goal of supporting additional input formats such as AsyncAPI, GraphQL, Protobuf or Cap’n Proto has not yet been demonstrated, and the effectiveness of chaining transformers across diverse languages is still speculative. The impact on API versioning strategies, including the upcoming shift from the v4 API that has been stable for 10 years, has not been quantified.