Weft is a young open-source programming language with 2,000+ stars on GitHub, designed from scratch for one specific job: expressing AI orchestrations — the branching, parallel, model-calling logic that glues modern AI systems together.
Most agent systems today are written in general-purpose languages, with orchestration logic smeared across prompt strings, callback chains and framework boilerplate. Weft’s bet is that this logic deserves a language of its own: a small, purpose-built language in which composing models, tools and control flow is the native construct rather than an afterthought. It is an early-stage project — the kind of repository whose star count reflects curiosity about the idea as much as production adoption — and it is evolving quickly, with updates landing in October 2026. That makes it interesting on two levels: practically, as a tool to experiment with if you build multi-step AI pipelines and feel the friction of doing so in Python or TypeScript; and conceptually, as a statement about where AI engineering may head if orchestration keeps growing in complexity. Documentation lives in the repository, and, as with any young language, expect APIs and syntax to keep moving.
Weft is free and open source. Programs you write in it will typically call models whose providers charge their own rates, but the language itself has no licensing cost or paid tier.
AI engineers and language-curious developers who build complex orchestrations and want to see what a language designed specifically for that job feels like — early adopters comfortable with a fast-moving, experimental project.