Javascript SDK

Our Javascript SDK is a wrapper around our Rest API. You can use it to integrate the workflows you build into your NodeJS codebase.

Setup the SDK

You can install the sdk using npm

npm i @pufflig/ps-sdk

or yarn

yarn add @pufflig/ps-sdk

Usage

You can quickly start making requests with the following code snippet:

import { runInstruction } from "@pufflig/ps-sdk";

const { datapoint } = await runInstruction({
  apiKey: YOUR_PROMPT_STUDIO_API_KEY,
  modelId: "gpt-3.5-turbo-instruct",
  prompt: "what is the fastest car?",
  parameters: {
    temperature: 0,
  },
});

console.log(datapoint.modelOutput); // As of 2021, the fastest car in the world is the Bugatti Chiron Super Sport 300+, with a top speed of 304 mph.

Your API key is located in Prompt Studio, specifically under the "API" tab in the tool editor. Note that each workspace within Prompt Studio has a distinct key.

Datapoints

Datapoints track completions you or your users created, these datapoints can later be used to finetune LLMs to better serve your problem.

Available Models

• gpt-3.5-turbo-instruct
• gpt-4-1106-preview
• meta-llama/llama-2-13b-chat
• anthropic/claude-2

Each model accepts different parameters.

API Reference

runInstruction(RunInstructionInput)

Generate a completion given a prompt. If the prompt is larger than the context length permitted by the LLM the start and end of the prompt are preserved while the center is summarized.

Input

name	type	data type	description
apiKey	required	string	Your Prompt Studio api key
modelId	required	string	The model to call
prompt	required	string	The prompt to provide to the language model
parameters	optional	Record<string, string>	The parameters to provide to the language model
options	optional	Options	Request options

Return

name	data type
datapoint	Datapoint / undefined

runWorkflow(RunWorkflowInput)

Run a deployed workflow on Prompt Studio.

Inputs

name	type	data type	description
apiKey	required	string	Your Prompt Studio api key
deploymentId	required	string	Your Prompt Studio api key
input	required	Record<string, any>	Input to a workflow
options	optional	Options	Request options

Return

name	data type
datapoint	Datapoint / undefined

Datapoint

name	type	data type	description
modelOutput	optional	string	Output of the language model
modelInput	optional	string	Input to the language model
workflowInput	optional	Record<string, any>	Input to a workflow
workflowOutput	optional	Record<string, any>	Output of a workflow
modelId	optional	string	Id of the model used

Options

name	type	data type	description
cache	optional	boolean	Return cached value if true, note some parameters like temperature will invalidate cache
store	optional	boolean	If true, store the completion in Prompt Studio for later use, e.g. for fine-tuning