Prompts

What it is

A Prompt is the core resource in PromptOps — a named, versioned piece of text that your AI application fetches at runtime instead of hard-coding it.

Instead of writing prompt text inside your code and redeploying every time it changes, you store it here and fetch it by key. Editors can change the wording; your app picks it up without a release.

Anatomy of a prompt

Field	Description
key	Stable identifier the SDK fetches by, e.g. sql.generator
name	Human-readable label
content	The prompt text, with placeholders
env	Environment — dev, test or prod
version	The current working version label, e.g. v1
category	Free-text grouping, e.g. code
tags	Comma-separated tags, e.g. review, sql
model	Suggested model for this prompt, e.g. gpt-4o

The same key can exist once per environment — sql.generator in dev and sql.generator in prod are two separate prompts.

Template variables

Prompt content can contain placeholders. They are filled in at render time — by the SDK's render(), by the Playground, or by an Agent:

You are a {{language}} expert.
Please review the code below:
{{code}}

The editor automatically detects every in the content and shows them, so you always know what inputs a prompt expects.

Using it

1. Open the Prompts page and click New Prompt.
2. Fill in key, environment, content and any optional metadata.
3. Click Save.
4. Use the search box (matches key / name / content) and the environment filter to find prompts later.

Fetching a prompt from your app

import { PromptOpsClient } from '@promptops/client'

const client = new PromptOpsClient({ server: 'http://localhost:8080', namespace: 'prod' })

// renders {{variable}} placeholders with the values you pass
const text = await client.render('sql.generator', { question: 'list all users' })

Next steps

• Saving a prompt is editable and mutable. To freeze it, see Versioning & Publishing.
• To push changes to running apps instantly, see Hot Reload.