@premai/prem-sdk

Prem Javascript SDK

Installation

You can install the Prem Javascript SDK directly from npm.

npm install @premai/prem-sdk

Usage

Getting Started

To use the Prem Javascript SDK, you need to obtain an API key from the Prem platform. You can then create a Prem instance to make requests to the API.

import Prem from '@premai/prem-sdk';

const client = new Prem({
  apiKey: "YOUR_API_KEY"
})

const project_id = PROJECT_ID

Chat completion

The chat.completions module allows you to generate completions based on user input.

Note that system is NOT an acceptable role: to use a system prompt you should define and pass system_prompt.

Here's an example:

const system_prompt = "You're an helpful assistant"
const messages = [
    { role: "user", content: "Who won the world series in 2020?" },
]

// Create completion
const responseSync = await client.chat.completions.create({
  project_id,
  system_prompt,
  messages
})

console.log(responseSync)

Chat completion with stream

You can also create a completion with a stream to receive the response in chunks by passing the stream parameter as true (default is false).

// Create completion with stream
const responseAsync = await client.chat.completions.create({
  project_id,
  messages,
  stream: true
})

for await (const chunk of responseAsync) {
  if (chunk.choices[0].delta.content) {
    process.stdout.write(chunk.choices[0].delta.content)
  }
}

console.log("\nTrace ID", responseAsync.trace_id)

Prompt Templates

Should your operations entail the frequent utilization of identical prompt structures, the Prompt Template functionality facilitates the streamlining of this process. It enables the instantiation and subsequent reuse of predefined prompts, thereby optimizing efficiency and maintaining uniformity across interactions.

Creating a Prompt Template

Create your own Prompt Template in just a few steps:

• Navigate to the Launchpad section of your project.
• Click on the Templates tab.
• Hit the button to create a new Prompt Template.

From here, you can either create your custom Prompt Template or choose one of our default presets.

Within the template, you can include placeholders for dynamic content by using the ${placeholder_name} syntax, as illustrated below:

Summarize the following text:
"""
${text}
"""

In this example, we have a placeholder named text. To implement this through our SDK, follow this sample code:

// Text you want to summarize
const text_to_summarize = "This is the great tale of ... "
// Construct the message with the template
const messages = [
    {
        role: "user",
        template_id: TEMPLATE_ID,  // Your template's ID
        params: {text: text_to_summarize}
    }
]

const response = await client.chat.completions.create({
    project_id,
    messages
})

Key Points for Using Prompt Templates

When using prompt templates, remember these important guidelines:

• Replace the content field with template_id and params.
• Both template_id (the unique ID of your prompt template) and params (a key-value object mapping your placeholders to their desired values) are required to utilize prompt templates.

Any keys in params not matching placeholders will be ignored. If a placeholder is omitted in params, it defaults to an empty string. For instance, if you provide the following message set, the ${text} placeholder will be left empty:

const messages = [
    {
        role: "user",
        template_id: TEMPLATE_ID,
        params: {}  // No parameters provided for placeholders
    }
]

Optional parameters

By default, the chat.completions module uses the default launchpad parameters. You can also specify the following optional parameters:

• model: The model to use for completion. If omitted, the default launchpad model will be used.
• system_prompt: The system prompt to use for completion. If omitted, the default launchpad system prompt will be used.
• session_id: A unique identifier to maintain session context, useful for tracking conversations or data across multiple requests.
• temperature: The temperature to use for completion. If omitted, the default launchpad temperature will be used.
• max_tokens: The maximum number of tokens to generate for completion. If omitted, the default launchpad max tokens will be used.

Example:

const model = "gpt-3.5-turbo"
const system_prompt = "You are a helpful assistant."
const session_id = "my-session"
const temperature = 0.7
const messages = [
    { role: "user", content: "Who won the world series in 2020?" },
]

// Create completion
const response = await client.chat.completions.create({
  project_id,
  messages,
  model,
  system_prompt,
  session_id,
  temperature
})

console.log(response)

Enhanced Chat Completion with Retrieval Augmented Generation (RAG)

Enhance your chat completions by leveraging contextual data from specified repositories. A repository is a collection of documents, each containing information that can be utilized by the RAG system to provide enriched and context-aware responses.

If you've linked your repositories in the launchpad, relax—you're all set for effortless chat completions! The system automatically uses those parameters by default, ensuring a seamless and easy experience. However, if you wish to customize the process, you can specify the repositories parameter to fit your exact needs. Just define:

• ids: Your selected repository IDs.
• similarity_threshold: The least similarity score for content relevance.
• limit: The number of content pieces to include.

For guidance on managing repositories, see the Repositories section.

const messages = [
    { role: "user", content: "Which is Jack's pet name?" },
]

const repositories = {
  ids: [REPOSITORY_ID, ...],
  similarity_threshold: 0.65,
  limit: 3
}

// Create completion
const response = await client.chat.completions.create({
  project_id,
  messages,
  repositories,
  stream: false
})

console.log(response.choices[0].message.content)
// "Jack's pet name is Sparky."

console.log(response.document_chunks)
// E.g. [{ repository_id=4, document_id=14, chunk_id=15, document_name="pets_and_their_owners.txt", similarity_score=0.67, content="..." }, ...]

Repositories

Repositories act as storage for documents, organized to facilitate efficient information retrieval. Manipulating repository content is straightforward.

Document creation

To add a document to a repository, you can use the create method provided by the document API. Here's an example of how to create and upload a document:

const FILE_PATH = "pets_and_their_owners.txt"
// Content: "My friend Jack has a beautiful pet, he gave it the name Sparky, [...]"

const response = await client.repository.document.create(
    REPOSITORY_ID,
    {
        file: FILE_PATH
    }
)

console.log(response)
// E.g. {repository_id: 4, document_id: 14, name: "pets_and_their_owners.txt", type: "text", status: "UPLOADED", chunk_count: 0, error: null}

After uploading, the document state is reflected in fields such as:

• status: Shows UPLOADED initially, changes once processed (e.g., PROCESSING).
• chunk_count: Number of data chunks; starts at 0 and increases post-processing.
• error: Non-null if an error arose during processing.

We currently support below file formats for document uploads:

• .txt
• .pdf
• .docx