mothertongue

A new take on AI-assisted programming in which you can write natural language and output working code in any language.

How does it work?

mothertongue will watch your files. When you make an update to any .mother file, it will be sent to an LLM of your choosing and translated to the language of your choosing.

Simply run: mothertongue . in your project root and it'll do its magic.

Note: Right now only Gemini is supported out-of-the-box. Set the following environment variable: GEMINI_TOKEN to a working AI Studio API Key to use mothertongue.

Workflow

Let's consider a scenario where I have the following directory structure:

/web_app/
    - models/
        - User.mother
    - package.json
    - node_modules/
    - index.mother

When mothertongue runs its translation process, all .mother files will be processed, creating an appropriate file for it. Consider index.mother has the following contents:

#!ts

#@ models/User

start:
- create User with username: "john"
#; output via stdout not console
- output user's username

You may notice some interesting syntax here. While mothertongue allows natural language, there is some syntactic sugar here that helps the LLM understand what you want.

.mother specifiers

#! specifier

#! is ALWAYS the first line of a .mother file. This determines what language something is written in. If code is version-specific, it's worth mentioning that here too. Examples:

• #!ts
• #!Go lang
• #!C# on .NET 10-preview

The purpose of #! is strictly to inform the LLM as to what kind of file it will be writing.

---

#@ specifier

#@ is completely optional, but you'll likely want to use this one. It informs the LLM of an import elsewhere. Example: #@ models/User. This tells the AI: "There's an import in models/User that we can have access to! Use it wisely.

The #@ specifier is special because when included in a .mother file, the AI is actually fed an entire, recursive directory tree of all files from the root project folder. Ensure this is only being used where necessary.

---

#; specifier

#; has no particular function. It's goal is to provide the LLM additional context. The idea here is that #; will not be treated as regular instruction, wherein the LLM will not accidentally translate this natural language comment as code. Example:

#; ensure we are not using console.log

output "hello world!"
output "hello mars!"
output "the moon has water!"