Harper Reed's 2389 Research has released Dippin, an open-source domain-specific language (a programming language designed for one specific task) built to replace the clunky Graphviz DOT files that many teams use to define AI pipeline workflows.
The problem Dippin solves is familiar to anyone who has tried to build multi-step AI workflows in DOT format: you spend more time fighting escaped characters and missing backslashes than actually designing your pipeline. One engineer on the team reportedly burned 40 minutes debugging a pipeline that broke because of a single missing backslash. Dippin replaces that with typed syntax that lets you write multi-line prompts and shell scripts without escaping, compose reusable subgraph patterns, and define branching logic with proper type checking.
The tooling around the language is where things get interesting. Dippin ships with 39 diagnostic checks for catching structural and semantic errors, scenario testing with deterministic path assertions, cost estimation that counts prompt tokens and applies per-model pricing (including retry loops), and LSP integration for real-time editor feedback. There is also an automated migration tool for converting existing DOT files.
Dippin compiles to a typed intermediate representation, so every tool in the chain works from the same structured format rather than re-parsing raw text. It is written in Go and installable with a single command.
This is a niche tool aimed squarely at teams building complex, multi-step LLM orchestration workflows. If your AI pipeline fits in a single API call, you do not need this. But for teams maintaining dozens of interconnected agents with branching logic and retry policies, the jump from escaped strings to a proper typed language is a practical improvement.