When your prompt is ready for production, you can easily integrate it into your system.

Asserto functions as a CMS for your prompts, separating prompt content from your application code. Once integrated, you can push prompt updates without the need to redeploy your system—keeping your production workflow lean and flexible.

​
1. Getting the Prompt Content (Messages and Configuration)

Prompt data includes the messages to be sent to LLM and configuration options like model name and temperature.

{
  "slug": "my-prompt",
  "messages": [
    {
      "role": "system",
      "content": {
        "type": "text",
        "text": "You are a sentiment classifier.\nClassify the following movie review as 'positive' or 'negative'.\nRespond in JSON format:\n{\"sentiment\": \"<result>\"}\n"
      }
    },
    {
      "role": "user",
      "content": {
        "type": "text",
        "text": "Review: {{review}}"
      }
    }
  ],
  "model": {
    "name": "gpt-4o-mini",
    "url": null,
  },
  "temperature": 0.1
}

You can retrieve this information in two ways:

​
Export as JSON via the UI

From the Prompt Page, click the Use Prompt button and select Export as JSON. You can also export all prompts at once from the prompt-list page.

​
Get Prompt via API

Generate a project API key from the Settings page.

Use the following endpoint to fetch the prompt definition:

curl -H "Authorization: ApiKey <YOUR_API_KEY>" https://app.asserto.ai/api/prompts/PROMPT_SLUG

Refer to the API documentation for more details.

​
2. Sample Code for Integration with OpenAI API

Once you have the prompt messages and configuration, you can render them and send the request using the OpenAI API.

Here’s an example in python using chevron, a Mustache templating library:

from openai import OpenAI
from chevron import renderer


# Set OPENAI_API_KEY. See https://platform.openai.com/docs/quickstart
client = OpenAI()

# TODO: Replace with values
data = {
    "review": None,
}

messages = [
    {
        "role": "system",
        "content": {"type": "text", "text": "You are a sentiment classifier.\nClassify the following movie review as 'positive' or 'negative'.\nRespond in JSON format:\n{\"sentiment\": \"<result>\"}\n"}
    },
    {
        "role": "user",
        "content": {"type": "text", "text": "Review: {{review}}"}
    },
]
rendered_messages = []
for msg in messages:
    msg['content'] = renderer.render(msg['content']['text'], data)
    rendered_messages.append(msg)

response = client.chat.completions.create(
    model="gpt-4o-mini",
    messages=rendered_messages,
    temperature=0.1,
)

By combining this with the prompt API, your system can stay in sync with content changes without code changes or redeployments.