Quickstart

This walkthrough gets you from zero to a typed prompt call in your code. We'll create a workspace, publish a tiny prompt, install the CLI, generate code, and call the prompt.

1. Sign up and create a workspace

Head to sufleur.com, register an account, and create a workspace. Your workspace name becomes the prefix for every prompt you publish: a workspace called acme produces prompts like @acme/welcome-message.

2. Generate an API key

Open your workspace settings and create an API key. Keep it somewhere safe — you'll feed it to the CLI through an environment variable in the next step.

3. Author and publish a prompt

In the web app, create a new prompt called welcome-message. Add a single entrypoint file (for example userPrompt) with this template:

Hi {{user.name}}{{@type string}}{{@doc User's first name}}, welcome aboard!

Save the draft, then publish version 0.1.0.

4. Install the CLI

Pick the wrapper that matches your project. Both ship the same Go binary.

pnpm add -D @sufleur/cli
# or: npm i -D @sufleur/cli
# or: yarn add -D @sufleur/cli

5. Initialize your project

From the root of your project, run:

sufleur init

You'll be prompted for your workspace name, the env var that holds your API key, the output language, and where to write the generated file. The result is a sufleur.yaml like this:

api_keys:
  acme: ${ACME_API_KEY}
 
prompts: {}
 
output:
  language: typescript
  file: ./generated/prompts.ts

Set ACME_API_KEY in your environment or in a local .env file (the CLI loads .env automatically).

6. Add the prompt and generate code

sufleur add @acme/welcome-message
sufleur generate

add validates the prompt exists, writes it to sufleur.yaml, resolves the latest matching version, and caches it under .sufleur/. generate reads the lockfile and writes typed code to the path in output.file.

7. Call the prompt from your code

import { getPrompt } from './generated/prompts'
 
const welcome = getPrompt('@acme/welcome-message')
 
const { prompt } = welcome.render('userPrompt', {
  user: { name: 'Ada' },
})
 
// `prompt` is the rendered string — pass it to your LLM client of choice.
console.log(prompt)

The user.name argument is required and typed — your editor knows the shape because the CLI inferred it from the Mustache template plus the {{@type string}} annotation.

What's next

• Add an output schema so parseOutput() returns a validated object.
• Read the CLI reference for update, --frozen (CI mode), and aliases.
• Open the SDK page for your language to see the full surface of the generated code.