prompt engineering, compiled

HowTo AI

A toolkit for people who treat prompts like engineering artifacts. Write structured scaffolds, version every change, and ship results you can reproduce.

Built in public. The Discord is where the recipes, benchmarks, and early builds land first.

4
scaffold fields
12+
model targets
beta
stage
howto compile

$ howto new "rewrite our API changelog"

role:senior technical writer

context:internal API docs, v2 migration

constraints:≤ 200 words, no marketing tone

output:markdown, h2 + bullet list

compiled prompt · 1 of 3 variants

ready

reproduciblediffableportable

Built like a tool, not a prompt dump

/ specs
01

Structured scaffolds

Role, context, constraints, output contract. Every prompt compiles down to a predictable schema instead of a wall of text.

role · context · constraints · output

02

Versioned iteration

Fork a prompt, diff the change, keep the variant that scored better. Your prompt library behaves like a codebase.

fork · diff · rollback

03

Model agnostic

One recipe, multiple targets. Adapters normalize tone, token budget, and formatting per model family.

gpt · claude · gemini · local

04

Private by default

Your prompts stay yours. No training on your data, no silent sharing, plain-language policies.

no training · no resale

Shipping in the open

/ roadmap
  1. Prompt engine

    shipped

    The scaffold compiler and schema behind every recipe.

  2. Recipe library

    in progress

    Community templates, tags, and scoring.

  3. Web workspace

    queued

    Browser editor with side-by-side model runs.

  4. Discord companion

    queued

    Generate and share prompts straight from chat.

  5. Public launch

    queued

    Open access, docs, and an API.

Get in before v1

Early access, prompt reviews, and a direct line to the people building it. The Discord is the whole roadmap in real time.

discord.gg/howto