AI agent stdlib that works with any LLM and TypeScript AI SDK.

• Intro
• Install
• Services
• Compound Tools
• AI SDKs
• Client Goals
• TODO
• Contributors
• License

The goal of this project is to create a set of standard AI functions / tools which are optimized for both normal TS-usage as well as LLM-based apps and that work with all of the major AI SDKs (LangChain, LlamaIndex, Vercel AI SDK, OpenAI SDK, etc).

For example, stdlib clients like WeatherClient can be used as normal TS classes:

import { WeatherClient } from '@agentic/stdlib'

const weather = new WeatherClient() // (requires `WEATHER_API_KEY` env var)

const result = await weather.getCurrentWeather({
  q: 'San Francisco'
})
console.log(result)

Or you can use these clients as LLM-based tools where the LLM decides when and how to invoke the underlying functions for you.

This works across all of the major AI SDKs via adaptors. Here's an example using Vercel's AI SDK:

// sdk-specific imports
import { openai } from '@ai-sdk/openai'
import { generateText } from 'ai'
import { createAISDKTools } from '@agentic/stdlib/ai-sdk'

// sdk-agnostic imports
import { WeatherClient } from '@agentic/stdlib'

const weather = new WeatherClient()

const result = await generateText({
  model: openai('gpt-4o'),
  // this is the key line which uses the `@agentic/stdlib/ai-sdk` adaptor
  tools: createAISDKTools(weather),
  toolChoice: 'required',
  prompt: 'What is the weather in San Francisco?'
})

console.log(result.toolResults[0])

You can use our standard library of thoroughly tested AI functions with your favorite AI SDK – without having to write any glue code!

Here's a slightly more complex example which uses multiple clients and selects a subset of their functions using the AIFunctionSet.pick method:

// sdk-specific imports
import { ChatModel, createAIRunner } from '@dexaai/dexter'
import { createDexterFunctions } from '@agentic/stdlib/dexter'

// sdk-agnostic imports
import { PerigonClient, SerperClient } from '@agentic/stdlib'

async function main() {
  // Perigon is a news API and Serper is a Google search API
  const perigon = new PerigonClient()
  const serper = new SerperClient()

  const runner = createAIRunner({
    chatModel: new ChatModel({
      params: { model: 'gpt-4o', temperature: 0 }
    }),
    functions: createDexterFunctions(
      perigon.functions.pick('search_news_stories'),
      serper
    ),
    systemMessage: `You are a helpful assistant. Be as concise as possible.`
  })

  const result = await runner(
    'Summarize the latest news stories about the upcoming US election.'
  )
  console.log(result)
}

Here we've exposed 2 functions to the LLM, search_news_stories (which comes from the PerigonClient.searchStories method) and serper_google_search (which implicitly comes from the SerperClient.search method).

All of the SDK adaptors like createDexterFunctions accept very flexible in what they accept. AIFunctionLike objects include:

• AIFunctionSet - Sets of AI functions (like perigon.functions.pick('search_news_stories') or perigon.functions or serper.functions)
• AIFunctionsProvider - Client classes which expose an AIFunctionSet via the .functions property (like perigon or serper)
• AIFunction - Individual functions (like perigon.functions.get('search_news_stories') or serper.functions.get('serper_google_search') or AI functions created directly via the createAIFunction utility function)

You can pass as many of these AIFunctionLike objects as you'd like and you can manipulate them as AIFunctionSet sets via .pick, .omit, .get, .map, etc.

npm install @agentic/stdlib

This package is ESM only and requires Node.js >= 18 or an equivalent environment (bun, deno, CF workers, etc).

Depending on the AI SDK and tool you want to use, you'll also need to install the required peer dependencies.

Note that many of these clients expose multiple AI functions.

• SearchAndCrawl

• OpenAI SDK
  • no need for an adaptor; use AIFunctionSet.specs or AIFunctionSet.toolSpecs
• Vercel AI SDK
  • import { createAISDKTools } from '@agentic/stdlib/ai-sdk'
• LangChain
  • import { createLangChainTools } from '@agentic/stdlib/langchain'
• LlamaIndex
  • import { createLlamaIndexTools } from '@agentic/stdlib/llamaindex'
• Firebase Genkit
  • import { createGenkitTools } from '@agentic/stdlib/genkit'
• Dexa Dexter
  • import { createDexterFunctions } from '@agentic/stdlib/dexter'

See the examples directory for examples of how to use each of these adaptors.

• clients should be as minimal as possible
• clients should use ky and zod where possible
• clients should have a strongly-typed TS DX
• clients should expose select methods via the @aiFunction(...) decorator
  • inputSchema zod schemas should be as minimal as possible with descriptions prompt engineered specifically for use with LLMs
• clients and AIFunctions should be composable via AIFunctionSet
• clients should work with all major TS AI SDKs
  • SDK adaptors should be as lightweight as possible and be optional peer dependencies of @agentic/stdlib

• services
  • browserbase
  • brave search
  • phantombuster
  • apify
  • perplexity
  • valtown
  • replicate
  • huggingface
  • skyvern
  • pull from langchain
    • provide a converter for langchain DynamicStructuredTool
  • pull from nango
  • pull from activepieces
  • general openapi support ala workgpt
• compound tools / chains / flows / runnables
  • market maps
• incorporate zod-validation-error
• investigate autotool
• investigate alt search engines
• investigate data connectors

• Travis Fischer
• Kevin Raheja
• David Zhang
• Philipp Burckhardt
• Riley Tomasek
• And all of the amazing OSS contributors!

MIT © Travis Fischer

To stay up to date or learn more, follow @transitive_bs on Twitter.