Jasper

Promptfree, for everyone, and the only version there is

Copy the prompt for Claude CodeCopy the prompt for CodexCopy the prompt for Cursor

Copy

Build me the copy tooling I actually need instead of Jasper — and notice what you are paying for, because it is not the model.

**It is templates and a brand voice.** The model underneath is one you can call directly, usually for a fraction of the seat price. What the product adds is a starting point for each format and a way to keep the tone consistent, and both are files.

STACK
- Node 20+ with Fastify, server-rendered HTML
- SQLite through better-sqlite3, WAL mode
- A model API, or a local model if the material must not leave the building
- Caddy in front

THE DATA MODEL
- voices: id, name, description, rules_md, examples_json — the brand voice, written down
- templates: id, name, format, prompt_md, inputs_json, voice_id
- generations: id, template_id, inputs_json, output, model, tokens_in, tokens_out, cost_cents, created_at, rating, edited_output
- assets: id, kind, body, tags_json — the product facts, the proof points, the things that must be said correctly
- Every generation stored with what it cost. That column is how you find out whether this is cheaper than the subscription, and it usually is by a lot

THE BRAND VOICE, WHICH IS THE ACTUAL WORK
- Not a slider. A written document: the words you use, the words you never use, sentence length, whether you address the reader as 'you', how you talk about competitors, and three examples of copy that is right
- **That document is the product**, and writing it is a morning with somebody who knows the brand. It works with any model and outlives all of them
- Applied by putting it in the prompt, not by fine-tuning. Fine-tuning is expensive, slow to iterate and worse at this than a good instruction

TEMPLATES
- One per format you actually produce: a product description, a launch email, a landing headline set, a social post
- Each is a prompt with named inputs and the voice attached
- Kept in files, versioned, reviewed. A template that improved because somebody edited it once and nobody knows what changed is a template that will drift back

THE PART THAT MATTERS MORE THAN GENERATION
- **Store the edited version beside the generated one.** The difference between what the model wrote and what shipped is the highest-quality training signal you will ever have, and it costs one column
- Feed a few of those edits back into the voice document. That loop is what makes the output stop needing heavy editing
- Rate outputs, and look monthly at which templates produce copy nobody changes

THE HONEST WARNINGS
- **Never publish generated copy unread.** Not because it is bad, but because it is confident about facts it invented — a price, a feature, a claim about a competitor. Put the product facts in as assets and check every claim against them
- If the copy is about a regulated subject, a person signs it off. That is not a workflow preference
- If you use a hosted model, your drafts and your product plans go to a third party. Say so in the README, and use a local model if that is unacceptable

WHAT MATTERS MOST
The voice document and storing the edit. The first is what makes output usable; the second is what makes it improve. The model is the cheap, replaceable part.