Lang Anchor - A mini AI prompt framework

Lang Anchor provides a structured way to define AI prompts, manage missions, enforce rules, and format output results.

Install

Install Lang Anchor via npm or yarn:

npm i @knfs-tech/lang-anchor
#or
yarn add @knfs-tech/lang-anchor

Features

• Prompting: Helps you build standard reminder forms, optimizing your reminder process by including usage role, AI role, tasks, rules, output data, and reinforcement data
• Rule Constants: Includes built-in rule set to increase prompt accuracy and efficiency.
• Language Constants: A set of available languages, convenient to use
• Roles Constants: A set of user and AI roles are available for ease of use
• Chain Construction: Build a list of steps to manipulate axiomatic data for RAG and AI Agents models

Basic Usage

Example with PromptTemplate

const { Mission, ResultForm, PromptTemplate } = require("@knfs-tech/lang-anchor")
const { rules, roles, languages } = require("@knfs-tech/lang-anchor").constants

const variableInputs = [
  { type: "string", name: "text" },
]

const tasks = [
  "Summarize the given text into key points: {text} ",
  "Translate the give text to Vietnamese: {text}",
]

const mission = new Mission(tasks, variableInputs)

const resultForm = new ResultForm("json", {
  summary: "string",
  keyPoints: "string",
  translation: "string"
})

const modelRole = roles.modelRole.TRANSLATOR

const ruleApplies = [
  rules.noMoreCreativity,
  rules.onlyFormatResult,
  rules.noBias,
  rules.factualOnly
]

const prompt = new PromptTemplate(
  mission,
  resultForm,
  {
    modelRole,
    rules: ruleApplies,
  }
)

const inputData = {text: "Lang Anchor is powerful LLM AI framework..."}

prompt.addEnhancedData("keyPoints should be defined by ,")

//get prompted data
const query = prompt.get(inputData)

Example with Chain

const { Mission, ResultForm, PromptTemplate, Chain } = require("@knfs-tech/lang-anchor")
const { rules, roles, languages } = require("@knfs-tech/lang-anchor").constants

const processChain = new Chain()

const inputData = {text: "Lang Anchor is powerful LLM AI framework..."}

processChain
  .addStep(
    async() => {
      const variableInputs = [
        { type: "string", name: "text" },
      ]

      const tasks = [
        "Summarize the given text into key points: {text} ",
        "Translate the give text to Vietnamese: {text}",
      ]

      const mission = new Mission(tasks, variableInputs)

      const resultForm = new ResultForm("json", {
        summary: "string",
        keyPoints: "string",
        translation: "string"
      })

      const modelRole = roles.modelRole.TRANSLATOR

      const ruleApplies = [
        rules.noMoreCreativity,
        rules.onlyFormatResult,
        rules.noBias,
        rules.factualOnly
      ]

      const prompt = new PromptTemplate(
        mission,
        resultForm,
        {
          modelRole,
          rules: ruleApplies,
        }
      )

      prompt.addEnhancedData("keyPoints should be defined by ,")
  
      return prompt
    }
  )
  .addStep(
    async() => {
      // Add more context data, by getting more information from vectorDatabase
      const prompt = processChain.results[0]
      const vectorDBResults = getDataFromVectorDB(inputData.text)
       for (const result of vectorDBResults) {
         prompt.addEnhancedData(result)
       }  
    }
  )
  .addStep(
  	async() => {
      // Add more context data, by getting more information from vectorDatabase
      const prompt = processChain.results[0]
  		const query = prompt.get(inputData)
  		return await callLLMMode(query);
  	}
  )

const result = await processChain.run();

API Reference

Mission

new Mission(tasks, variableInputs)

Create mission of AI

| Parameter | Type | Default | Description | Support Version | | --------------------- | ------- | ---------------------- | ------------ | ----------------- | | tasks | Array string | None | AI task work list. | >= 0.1.1 | | variableInputs | Array ({type: , name: }) | None | Variable input of the task. | >= 0.1.1 |

const { Mission } = require("@knfs-tech/lang-anchor")

const variableInputs = [
	{ type: "string", name: "mainKeyWords" },
	{ type: "string", name: "secondaryKeywords"}
]

const tasks = [
	"Write me an SEO standard article with main keyword: {mainKeyWords} and secondary keywords: {secondaryKeywords} ",
]

const mission = new Mission(tasks, variableInputs)

get(valueInputs)

Get mission prompt with value inputs

| Parameter | Type | Default | Description | Support Version | | ---------------- | ------------------------------------------------------------------------- | ------- | --------------------------- | --------------- | | valueInputs | Array ({ : <value of variable input}) | None | Value of variable input. | >= 0.1.1 |

Returns: String - The prompted string

const inputData = {
	mainKeyWords: "Logitech wireless mouse",
	secondaryKeywords: "Logitech GHUB, Logitech mouse",
}

const missionPrompt = mission.get(inputData)

ResultForm

new ResultForm(type, structure)

Create result form of AI

| Parameter | Type | Default | Description | Support Version | | ---------------- | ------------ | ------- | --------------------------- | --------------- | | type | String | "string" | Type of response received when communicating with LLM. | >= 0.1.1 | | structure | String | "" | Structure of the returned response. | >= 0.1.1 |

const { ResultForm } = require("@knfs-tech/lang-anchor")

const resultForm = new ResultForm("json", {
	content: "string",
	keywords: "string",
})

get()

Get result prompt with value inputs

Returns: String - The prompted string

const resultFormPrompt = resultForm.get()

PromptTemplate

new PromptTemplate(mission, resultForm, options)

Create prompt tempt of AI

| Parameter | Type | Default | Description | Support Version | | ----------- | ------ | ---------- | ------------------------------------------------------ | --------------- | | mission | Mission | | Mission of AI | >= 0.1.1 | | resultForm | ResultForm | | Result form of response AI | >= 0.1.1 | | options.modelRole | string | roleCons.modelRole.PERSONAL_ASSISTANT | The role of the AI. | >= 0.1.1 | | options.userRole | string | roleCons.userRole.NORMAL_USER | The role of user. | >= 0.1.1 | | options.language | string | languageCons.en | The language for the result. | >= 0.1.1 | | options.rules | Array | [] | A set of rules the AI must follow.| >= 0.1.1 |

const { PromptTemplate } = require("@knfs-tech/lang-anchor");
const { rules, roles, languages } = require("@knfs-tech/lang-anchor").constants;

const modelRole = roles.modelRole.MARKETING_EXPERT

const ruleApplies = [
	rules.onlyFormatResult,
	rules.noBias,
	rules.factualOnly,
	rules.useFormalTone,
	rules.followInstructions
]

const prompt = new PromptTemplate(
	mission,
	resultForm,
	{
		modelRole,
		rules: ruleApplies,
		language: languages.vi
	}
)

addEnhancedData(enhanceData)

Get prompt with value inputs

| Parameter | Type | Default | Description | Support Version | | ------------- | ------ | ------- | -------------------------------------------------- | --------------- | | enhanceData | string | None | The enhanced contextual data to increase accuracy. | >= 0.1.1 |

prompt.addEnhancedData("The article has 3H2.")
prompt.addEnhancedData("Articles need to be at least 1000 words.")

get(valueInputs)

Get prompt with value inputs

| Parameter | Type | Default | Description | Support Version | | ------------- | ------------------------------------------------------------- | ------- | ------------------------ | --------------- | | valueInputs | Array ({ : <value of variable input}) | None | Value of variable input. | >= 0.1.1 |

Returns: String - The prompted string

const inputData = {
	mainKeyWords: "Logitech wireless mouse",
	secondaryKeywords: "Logitech GHUB, Logitech mouse",
}

const query = prompt.get(inputData)

Chain

new Chain()

Create step-by-step AI processing.

Returns: Chain - This chain

addStep(step)

Get prompt with value inputs

| Parameter | Type | Default | Description | Support Version | | ------------- | ------ | ------- | -------------------------------------------------- | --------------- | | step | function | None | The function handle in step. | >= 0.1.1 |

Returns: Chain - This chain

const { Chain } = require("@knfs-tech/lang-anchor")

const processChain = new Chain()

processChain
	.addStep(
		async () => console.log("a")
	)
	.addStep(
		async () => console.log("b")
  )
	.addStep(
		async () => {
      return "c"
		}
	)

run()

Run process

Returns: Promise<*> - Final result in result list

const reulst = await processChain.run()

Author

• Kent Phung

Owner

• Knfs.,jsc

More

License

Lang Anchor is open-sourced software licensed under the MIT license.