Product - unit-text

🤌 Why unit-text?

If you’re a developer like us:

• You’re probably aware you should write more (either because your leadership asks for it, or because you’d like to be invited as a conference speaker, or <insert your own reason here>…)
• You don’t know what to write about.
• You have considered letting ChatGPT write blog posts for you.

If you tried the last option (or read anything on the Internet in the past 2 years) you would also know why it’s not really that good of an idea.

Sure, you may get a perfectly good post, but it won’t be your own.

unit-text applies the concept of unit tests to prose. You are the one writing, the LLM is just your critic.

First, you define a goal, the audience you have in mind, and what you wanted them to do differently after they read your post. Then, you start writing and iterating on your draft. unit-text gives you feedback: you run tests to validate whether your draft is going in the right direction.

An AI copy editor, you could say.

⚙️ Installation

We recommend uv to run the CLI.

# to run the CLI straight away
uvx unit-text --help

# or if you prefer to install it
uv tool install unit-text

You may of course also use pip to install the CLI - or pipx if you prefer to install it in an isolated environment.

pipx install unit-text

# ...or if you like to live on the edge
pip install unit-text

🔨 Usage

Requirements

Either:

• Ollama must be running locally;
• OLLAMA_HOST should point to an Ollama server.

# To generate a blog idea
unit-text ideate

# To validate the working draft
unit-text test <path-to-the-draft.md>

📝 Process

The ideation phase is where you define your blog idea. It looks something like this:

Your idea is now stored in a unit-text.json file, in the current directory.

You may run unit-text ideate again at any time to finetune it.

Now you can start writing, let’s say you’re writing in a file called draft.md.

At any time, you can run unit-text test draft.md to validate it. Initially, it may look something like this:

Keep iterating on your draft, based on feedback from unit-text, until it passes all tests: