Weft
A self-hosted AI orchestration framework that runs on local Kubernetes, checks workflow connections before execution, and supports human approvals.
Weft is a language and self-hosted framework for developers building AI workflows that combine models, APIs, databases and human decisions. It runs on a local Kubernetes cluster with Docker and provides a graph editor in VS Code. The runtime coordinates the work, while each model receives the context assigned to its part of the task.
The compiler checks input and output types, required connections and graph structure before execution, then builds the program into a native Rust binary. Workflows can include loops and components that keep communicating while other work continues. When a step needs a human answer or an external event, Weft saves its state and resumes after the response arrives. The graph editor displays execution history and the values passed between steps; Chrome and Firefox extensions handle human forms and approvals.
Tangle helps coding assistants such as Claude Code, Codex and Cursor build and test Weft programs. It can run individual steps, inspect results and preserve accepted examples for comparison after code changes. The runtime also manages container-based infrastructure, including Postgres with storage that survives restarts. Custom nodes use Rust, and the catalog covers messaging, web search, image and audio generation. The included WhatsApp support bot uses OpenRouter for model calls, so those requests go to an external service even though the runtime runs locally. Weft uses the O'Saasy License.
