feat: enable OpenAI models #307
Conversation
- Added OPENAI_API_KEY to the Env interface - Added OPENAI_API_KEY to package.json dependencies - Added @ai-sdk/openai to package.json dependencies - Added Prompts interface to prompts-interface.ts - Added Prompts implementation to anthropic-llm.ts and openai-llm.ts - Added LLMType enum to llm-selector.ts - Added selectLLM and getCurrentLLMType functions to llm-selector.ts - Added AnthropicLLM and OpenAILLM classes to anthropic-llm.ts and openai-llm.ts - Added getModel function to model.ts - Added streamText function to stream-text.ts - Updated chatAction function in api.chat.ts to use selectLLM and getCurrentLLMType
Refactor the OpenAI LLM implementation in the `openai-llm.ts` file. - Update the model selection logic to support both 'gpt-4o' and 'o1-mini' models. - Add conditional logic to handle different models and their respective prompts and options.
beejaysea
changed the title
|
Could you add changing the base url so you can use something like Deepseek? |
Sure, one certainly could do that. For my purposes, this is a stab at getting the initial support to be functional. I believe that the next refactor should include a more-robust model selector (possibly via UI), and way of handling more of the providers that Vercel's AI SDK supports: https://sdk.vercel.ai/providers/ai-sdk-providers For instance, I'd also like to be able to use Anthropic's models via GCP or OpenAI's models via Azure, or Groq with one of the Llama3 models, and a different system prompt... |
|
OpenRouter is much more favorable - https://openrouter.ai/docs/requests |
|
Hey! Thanks for the PR 🙏 I think for this repo we'd like to stick with Anthropic only. But if you want to fork this repo and add more providers then you can absolutely do that! |
This PR enables the use of gpt-4o, gpt-4o-mini, o1-preview, and o1-mini.
There are a few new environment variables:
LLM_TYPE=(anthropic|openai) // defaults to anthropic
OPENAI_MODEL=gpt-4o // one of the above, must be present, no default
OPENAI_API_KEY=sk... // OpenAI key here
.toAIStream() has been deprecated in the newer Vercel AI SDK, and did not work with the OpenAI LLM implementation. This has been updated to use .toDataStream(...)
Notes: