Building Blocks¶

Every prompt in Prompty is assembled from five building blocks, held together by a composite Prompt. This page explains each block, what it contributes, and how they combine.

1. Overview of the building blocks¶

The five building blocks are designed to be independent and reusable. You write a persona once and point at it from many prompts. You build a library of tones and pick from it the way a chef reaches for a spice rack. Only the composite Prompt ties everything together.

Block	Versioned	Default	Selection	Length
Task	Part of Prompt	Never stored alone	Required in every prompt	Up to 10,000 characters
Persona	Yes	Public	Zero or one	Title: 80 chars, Description: 600 chars
Tone	No	Public	Zero or many	Label: 80 characters
Output	No	Public	Zero or one	Label: 80 characters
Constraint	No	Public	Zero or many	Text: 300 characters
Prompt (composite)	Yes	Public	The final saved artifact	Compiled prompt: 50,000 characters

Only Prompts and Personas are versioned. Tones, Outputs, and Constraints are edited in place: if you change a tone, the change is immediate and is reflected everywhere the tone is displayed. Saved prompts that referenced the old text of the tone keep a snapshot of that text, so their history stays intact.

2. Task¶

The Task is the main instruction of the prompt. It is the thing you actually want the model to do. Unlike the other blocks, Task is not a reusable entity: each prompt has its own task, typed fresh. You can copy a task from one prompt to another, but there is no Task library to browse.

Length: up to 10,000 characters. That is long enough for detailed rubrics, multi-step instructions, and example-driven tasks.
Improve button: clicks an AI pass that rewrites your task for clarity. It is subject to a monthly quota that depends on your plan.
Required: a task with no text produces a prompt with no task line. You can still save it, but you probably should not.

In the compiled prompt:

Your task: {task}

3. Persona¶

A Persona is the role or perspective the model should adopt. It is where you write things like "a senior technical editor who respects reader time" or "a patient algebra tutor explaining ideas to a fifteen-year-old." Good personas carry a lot of weight: a model that knows who it is behaves differently from one starting cold.

Title: up to 80 characters. A short human-readable label you will see in lists.
Description: up to 600 characters. The actual persona text that gets inserted into prompts.
Versioned: yes. Every edit creates an immutable new version with an optional changelog.

In the compiled prompt:

You are {persona description}.

4. Tone¶

Tones are the emotional and stylistic qualities you want in the response. Pick several. A tone can be as plain as "concise" or as specific as "wry without being cynical."

Label: up to 80 characters. Displayed as a pill you toggle on and off.
Selection: multi-select. You can pick any number of tones, including none.
Versioned: no. Edits apply in place. Saved prompts that referenced the old text keep it in their own snapshot.

In the compiled prompt:

The tone of the output should be {tones joined by ", "}.

5. Output¶

The Output block describes the shape of the response: a bulleted list, a numbered checklist, a markdown table, a JSON object, a polite email. It is the thing a downstream parser, reader, or tool expects.

Label: up to 80 characters.
Selection: single-select. A prompt has at most one output format.
Versioned: no. Edits apply in place.

In the compiled prompt:

The output format should be {output}.

6. Constraint¶

Constraints are the rules and limits the response must respect. Word counts, forbidden phrases, cited sources, reading levels. A good constraint is specific enough to be checkable and general enough to be reused.

Text: up to 300 characters. Long enough to explain nuance without being a second task.
Selection: multi-select. Stack as many as you need.
Versioned: no. Edits apply in place.

In the compiled prompt:

Always adhere to the following constraints: {constraints joined by ", "}.

7. Prompt¶

A Prompt is the composite artifact you save. It is a header with ownership and visibility, plus one or more versions that hold the actual content. Every time you edit a saved prompt, you create a new version. The old version stays available, exactly as it was.

Title: up to 200 characters.
Description: up to 600 characters.
Compiled prompt: up to 50,000 characters. This is the final assembled text that you would paste into an AI tool.
Changelog per version: up to 300 characters. A short note describing what changed.
References: each version stores links back to the persona, tones, output, and constraints it was built from, so you can trace a prompt to its ingredients.

Prompts are public by default. You can make a prompt private at any time, which adds it to your private-prompt budget on your tier.

8. How the prompt is assembled¶

Prompty assembles your compiled prompt from the building blocks using a fixed template. Each part is only included if you provided it. The pieces are joined with blank lines.

The assembly template:

You are {persona}.

Your task: {task}

The tone of the output should be {tones, joined by comma}.

The output format should be {output}.

Always adhere to the following constraints: {constraints, joined by comma}.

Here is a worked example. Suppose you picked this persona, this task, these two tones, this output, and these two constraints:

Persona: a supportive career counselor who specializes in tech transitions
Task: Help a career-switcher decide between DevOps and Data Engineering roles.
Tones: encouraging, informative
Output: Step-by-step guide
Constraints: Keep response under 500 words, Cite at least one industry source

Prompty compiles those into:

The compiled prompt:

You are a supportive career counselor who specializes in tech transitions.

Your task: Help a career-switcher decide between DevOps and Data Engineering roles.

The tone of the output should be encouraging, informative.

The output format should be Step-by-step guide.

Always adhere to the following constraints: Keep response under 500 words, Cite at least one industry source.

You can copy that text into any LLM tool. If you save the prompt, this exact compiled text is stored on the version along with the IDs of the blocks that produced it.

9. Why snapshots matter¶

Every saved prompt version stores the exact text of each building block as it was the moment you saved. This is intentional. If you improve a persona later, your earlier prompts do not suddenly change. The history of your work stays true.

Versions also store the IDs of the blocks they were built from. That means you can trace a prompt back to its ingredients and see the latest version of each. Snapshots protect the past; references connect you to the present.

A small example

You save a prompt in March that uses the tone "concise". In April you rename that tone to "lean". Your March prompt still says "concise" in its compiled text. New prompts you save after the rename will say "lean".

For how to manage, edit, and share these saved prompts, head to Your Library.