@markprompt/core

@markprompt/core

@markprompt/core is the core library for Markprompt, a conversational AI component for your website, trained on your data.

It contains core functionality for Markprompt and allows you to build abstractions on top of it.

Installation

npm install @markprompt/core

In browsers with esm.sh:

<script type="module">
  import { submitPrompt } from 'https://esm.sh/@markprompt/core';
</script>

Usage

import { submitPrompt } from '@markprompt/core';

// User input
const prompt = 'What is Markprompt?';
// Can be obtained in your project settings on markprompt.com
const projectKey = 'YOUR-PROJECT-KEY';

// Called when a new answer chunk is available
// Should be concatenated to previous chunks
function onAnswerChunk(chunk) {
  // Process an answer chunk
}

// Called when references are available
function onReferences(references) {
  // Process references
}

// Called when submitPrompt encounters an error
const onError(error) {
  // Handle errors
}

// Optional parameters, defaults displayed
const options = {
  model: 'gpt-3.5-turbo', // Supports all OpenAI models
  iDontKnowMessage: 'Sorry, I am not sure how to answer that.',
  apiUrl: 'https://api.markprompt.com/v1/completions', // Or your own completions API endpoint
};

await submitPrompt(prompt, projectKey, onAnswerChunk, onReferences, onPromptId, onError, options);

API

submitPrompt(prompt, projectKey, onAnswerChunk, onReferences, onError, options?)

Submit a prompt to the Markprompt Completions API.

Arguments

• prompt (string): Prompt to submit to the model
• projectKey (string): Project key for the project
• onAnswerChunk (function): Answers come in via streaming. This function is called when a new chunk arrives
• onReferences (function): This function is called when receiving the list of references from which the response was created.
• onError (function): called when an error occurs
• options (SubmitPromptOptions): Optional parameters

Options

• apiUrl (string): URL at which to fetch completions
• iDontKnowMessage (string): Message returned when the model does not have an answer
• model (OpenAIModelId): The OpenAI model to use
• systemPrompt (string): The prompt template
• temperature (number): The model temperature
• topP (number): The model top P
• frequencyPenalty (number): The model frequency penalty
• presencePenalty (number): The model present penalty
• maxTokens (number): The max number of tokens to include in the response
• sectionsMatchCount (number): The number of sections to include in the prompt context
• sectionsMatchThreshold (number): The similarity threshold between the input question and selected sections
• signal (AbortSignal): AbortController signal

submitChat(prompt, projectKey, onAnswerChunk, onReferences, onError, options?)

Submit a prompt to the Markprompt Completions API.

Arguments

• messages (ChatMessage[]): Chat messages to submit to the model
• projectKey (string): Project key for the project
• onAnswerChunk (function): Answers come in via streaming. This function is called when a new chunk arrives
• onReferences (function): This function is called when receiving the list of references from which the response was created.
• onError (function): called when an error occurs
• options (SubmitChatOptions): Optional parameters

Options

• apiUrl (string): URL at which to fetch completions
• iDontKnowMessage (string): Message returned when the model does not have an answer
• model (OpenAIModelId): The OpenAI model to use
• systemPrompt (string): The prompt template
• temperature (number): The model temperature
• topP (number): The model top P
• frequencyPenalty (number): The model frequency penalty
• presencePenalty (number): The model present penalty
• maxTokens (number): The max number of tokens to include in the response
• sectionsMatchCount (number): The number of sections to include in the prompt context
• sectionsMatchThreshold (number): The similarity threshold between the input question and selected sections
• signal (AbortSignal): AbortController signal

Returns

A promise that resolves when the response is fully handled.

submitSearchQuery(query, projectKey, options?)

Submit a search query to the Markprompt Search API.

Arguments

• query (string): Search query
• projectKey (string): Project key for the project
• options (object): Optional parameters

Options

• apiUrl (string): URL at which to fetch search results
• limit (number): Maximum amount of results to return
• signal (AbortSignal): AbortController signal

Returns

A list of search results.

Community

• Twitter
• Discord

Authors

This library is created by the team behind Markprompt (@markprompt).

License

MIT © Markprompt