dosco / llm-client Goto Github PK

• I'm submitting a ...
  [ ] bug report
  [X ] feature request
  [ ] question about the decisions made in the repository
  [ ] question about how to use this project

• Summary

Looking to add custom providers but immediately
looking for AWS Bedrock provider. Possible to add that?

This helps to integrate with new providers easily. Looking forward for an update soon.

Dosco, in the examples folder you have an ask-questions.js file mentioned in the readme but the actual file example is not linked. This the file that shows how to use Google Search with your library. Any chance you can add it?

• I'm submitting a ...
  [ ] bug report
  [X] feature request
  [ ] question about the decisions made in the repository
  [ ] question about how to use this project

• Summary
  Use Github discussions to start a discussion listing business workflow ideas we should target in this library. For example PII detection, Classification list for customer communications, Email communication rewriting, etc

• I'm submitting a ...
  [ ] bug report
  [ X] feature request
  [ ] question about the decisions made in the repository
  [ ] question about how to use this project

• Summary
  Build a new prompt to extract customer and product information. And to classify the customer communication.

• I'm submitting a ...
  [ ] bug report
  [X ] feature request
  [ ] question about the decisions made in the repository
  [ ] question about how to use this project

• Summary

We have a special AI client that customizes existing AI like Anthropic, Bedrock and OpenAI. In order to use llm-client with them, it helps to build on top of llm-client and requesting to make it possible.

1. BaseAI - https://github.com/dosco/llm-client/blob/main/src/ai/base.ts#L43 OR
2. a better alternative.

I'm submitting a bug report

• Summary
  The following code causes the exception: Cannot read properties of undefined (reading 'map'), this happens when calling a.forward

const ai = new Anthropic({
  apiKey,
});
const cot = new ChainOfThought(
  ai,
  `question:string -> answer:string`
);
const res = await a.forward({ question });

There is a lot of work that's happening around llm outputs. Some interesting things we can look at:

https://hwchase17.github.io/langchainjs/docs/modules/prompts/output_parsers/

https://github.com/shreyar/guardrails

https://eyurtsev.github.io/kor/index.html

We can look at specific implementations that are described here or we can build one from scratch.

Hey guys,

First I wanted to start by thanking the repo owner for including example scripts and samples in the README. However, I was wondering if there is any link to documentation that describes all the methods and classes available. This would be helpful in understanding how to leverage the package for novel workflows that have not been covered by the samples yet.