[Feature] llm Options

[729993031]

Background

I wish to experiment with other LLM models like Tongyi or Gemini, but it's not feasible.

stagehand/lib/llm/LLMProvider.ts

Line 62 in fecd42e

switch (provider) {

Aspiration

Provide a solution akin to browser-use, allowing the integration of custom models.

import { OpenAI } from "@langchain/openai";

const llm = new OpenAI({
  model: "gpt-3.5-turbo-instruct",
  temperature: 0,
  maxTokens: undefined,
  timeout: undefined,
  maxRetries: 2,
  apiKey: process.env.OPENAI_API_KEY,
  // other params...
});
await stagehand.init({llm});

Benefits

1. less code、lower maintenance complexity
2. The token cost of GPT-4 is relatively high; in certain scenarios, cheaper models can be employed to achieve similar outcomes.

[conradkoh]

+1, this would also be nice if we could use with OpenRouter for many reasons (e.g. rate limits, consolidated billing, experimenting with different models to determine the output quality against cost)

[achris7]

LiteLLM is another option. I second the request for gemini and other model options.

[cmaycumber]

This would be a great addition. We’re price sensitive it would help a lot.

[mugnimaestra]

up for this, i wish to use deepseek API which has cheaper price compared the current option with roughly similar capabilities

[ahlstro]

I would also like to be able to use DeepSeek, as its open source and can be downloaded. Should be one of the defaults togheter with the other two