Skip to main content

API

PromptDX supports 3 basic methods for interacting with prompts: runInference, deserialize, and serialize.

runInference​

Run inference will take a prompt config, call the model provider's API, and return the result in PromptDX format.

Usage​

import { runInference } from '@puzzlet/promptdx';
...
const result = await runInference(prompt, props)

Example:​

my-prompt.mdx:

---
name: my-prompt
metadata:
  model:
    name: gpt-4o-mini
---

<User>
Say: {props.greeting}
</User>

Run:

const props = { greeting: 'Hello' };
runInference(myPrompt, props);

Result:

Hello

deserialize​

Deserialize takes a prompt in PromptDX format, and turns it into the configuration needed to call a model provider.

Usage​

import { deserialize } from '@puzzlet/promptdx';
...
deserialize(prompt, props)

Example​

my-prompt.mdx:

---
name: my-prompt
metadata:
  model:
    name: gpt-4o-mini
---

<User>
Say: {props.greeting}
</User>

Run:

const props = { greeting: 'Hello' };
deserialize(myPrompt, props);

Result:

{
  "model": "gpt-4o-mini",
  "messages": [
    {
      "content": "Say: Hello",
      "role": "user",
    }
  ]
}

serialize​

Serialize allows you to take an existing configuration for a Prompt, and serialize it into a PromptDX file.

Usage​

import { serialize } from '@puzzlet/promptdx';
...
serialize(config, modelName, promptName)

Example​

Input Config:

{
  "model": "gpt-4o-mini",
  "top_p": 1,
  "temperature": 0.7,
  "messages": [
    {
      "content": "Say: {props.greeting}",
      "role": "user",
    }
  ]
}

Run:

serialize(config, "gpt-4o-mini", "basic-prompt")

Result (serialized string):

---
name: my-prompt
metadata:
  model:
    name: gpt-4o-mini
    settings:
      top_p: 1
      temperature: 0.7
---

<User>
Say: {props.greeting}
</User>